dynamic pricing in the presence of social learning and ...faculty.london.edu/nsavva/ps2014.pdf2...
TRANSCRIPT
Dynamic Pricing in the Presence of Social Learningand Strategic Consumers
Yiangos Papanastasiou · Nicos SavvaLondon Business School, Regent’s Park, London NW1 4SA, UK
[email protected] · [email protected]
When a product of uncertain quality is first introduced, consumers may be enticed to strategically delay
their purchasing decisions in anticipation of the product reviews of their peers. This paper investigates how
the presence of social learning interacts with the adoption decisions of strategic consumers and the dynamic-
pricing decisions of a monopolist firm, within a simple two-period model. When the firm commits to a price
path ex ante (pre-announced pricing), we show that the presence of social learning increases the firm’s ex
ante expected profit, despite the fact that it exacerbates consumers’ tendency to strategically delay their
purchase. As opposed to following a price-skimming policy which is always optimal in the absence of social
learning, we find that, for most model parameters, the firm will announce an increasing price plan. When
the firm does not commit to a price path ex ante (responsive pricing), interestingly, the presence of social
learning has no effect on strategic purchasing delays. Under this pricing regime, social learning remains
beneficial for the firm and prices may either rise or decline over time, with the latter being ex ante more likely.
Furthermore, we illustrate that contrary to results reported in existing literature, in settings characterized
by social learning, price-commitment is generally not beneficial for a firm facing strategic consumers.
Key words : Bayesian social learning, strategic consumer behavior, applied game theory, dynamic pricing
History : January 7, 2014
1. Introduction
The term “strategic consumer” is commonly used in the literature to describe a rational and
forward-looking consumer, who makes intertemporal purchasing decisions with the goal of maximiz-
ing her utility. In its simplest form, strategic behavior may manifest as bargain-seeking behavior,
whereby even if the current price of a product is lower than the customer’s willingness to pay, she
may delay her purchase in anticipation of a future markdown.1 The importance of forward-looking
consumer behavior in shaping firms’ pricing decisions has been widely recognized by practitioners
and academics alike: to defend against the negative effects of this behavior, firms are investing
heavily in price-optimization algorithms (e.g., Schlosser 2004), while the literature has produced
a number of managerial insights regarding how firms should adjust their approach to dynamic
1 See Li et al. (2013) for empirical evidence of strategic consumer behavior in the air-travel industry.
1
2 Author: Dynamic Pricing with Social Learning
pricing (e.g., Aviv and Pazgal 2008, Cachon and Feldman 2013, Mersereau and Zhang 2012, Su
2007). Apart from pricing, the effects of strategic consumer behavior also extend to a range of other
operational decisions; examples include decisions pertaining to stocking quantities (Liu and van
Ryzin 2008), inventory display formats (Yin et al. 2009), the implementation of quick-response and
fast-fashion practices (Cachon and Swinney 2009, 2011), the timing of new product launches (Bes-
bes and Lobel 2013), and the employment of advance selling (YU et al. 2013a), to name but a few.
Although existing research examines strategic consumer behavior from a variety of perspectives,
it generally does not account for cases in which the quality of a new product is ex ante uncertain
and, more importantly, for the prominent role of social learning (SL) in resolving this uncertainty.
In reality, many new product introductions are accompanied by quality uncertainty, in particular
owing to the ever-increasing complexity of product features. Examples of such products include
high-tech consumer electronics (e.g., smart-phones, tablets, computers), media items (e.g., movies,
books), and digital products (e.g., computer software, smart-phone apps). In the post-Internet era,
online platforms hosting buyer-generated product reviews offer a cheap and straightforward way of
reducing quality uncertainty; learning from buyer reviews has now become an integral component
of the process by which consumers make their purchasing decisions (e.g., Chevalier and Mayzlin
2006). For the consumers, learning from reviews allows for better-informed purchasing decisions,
which in turn reduces the likelihood of ex post negative experiences. For the firm, the SL process
can also be beneficial, for instance, by allowing for increased accuracy in forecasting future demand
(e.g., Dellarocas et al. 2007). However, the ease with which the modern-day consumer can gain
access to buyer reviews also gives rise to a new dimension of strategic consumer behavior: rather
than experimenting with a new product themselves, consumers may be enticed to delay their
purchasing decisions in anticipation of peer reviews. As a result, both the learning process (in
terms of information generation) as well as the firm’s performance (in terms of product adoption
and profit) may be significantly hampered.
Despite the well-documented importance of managing strategic consumer behavior, our under-
standing of the effectiveness of alternative operational decisions in settings characterized by SL is
extremely limited. In this paper, we take a first step towards developing such an understanding by
considering the fundamental problem of uncapacitated dynamic pricing. In particular, we set out to
address three research questions. First, how is the behavior of rational forward-looking consumers
altered by the opportunity for SL from buyer reviews? More specifically, we aim to gain insight
into how the SL process enters consumers’ adoption decisions, and identify the relative differences
in consumer behavior that result from alternative dynamic-pricing strategies. Second, given the
nature of strategic consumer behavior, how should profit-maximizing firms modify their pricing
decisions to accommodate the SL process? Here, we are interested in identifying the implications
Author: Dynamic Pricing with Social Learning 3
of SL for the optimal implementation of dynamic pricing, and explaining the underlying drivers of
these implications. Third, should firms facing strategic consumers commit to a price path ex ante
or adjust prices dynamically in time? The issue of price-commitment is one that arises frequently
in the strategic consumer literature. In general, the consensus is that price-commitment may prove
beneficial for the firm when consumers are forward-looking (e.g., Aviv and Pazgal 2008); it is of
interest to see whether and under which circumstances this finding applies to SL settings.
The model setting considered in this paper is much in the spirit of the seminal paper by Besanko
and Winston (1990). There is a monopolist firm selling a new product to a fixed population of
strategic consumers, over two periods. Two alternative classes of dynamic-pricing policies may
be employed: the firm may either (a) announce the full price path from the beginning of the
selling horizon (pre-announced pricing) or (b) announce only the first-period price, and delay the
second-period price announcement until the beginning of the second period (responsive pricing).2
Consumers are heterogeneous in their preferences for the product and make adoption decisions
to maximize their expected utility. Our addition to this simple model, and the focal point of our
analysis, is the introduction of ex ante quality uncertainty (faced by both the firm and consumers),
which may be partially resolved in the second period by observing the product reviews of first-
period buyers (SL).
Because in the presence of SL the product’s quality is partially revealed in the second period,
the interaction between the firm and consumers is transformed from a game whose outcome can
be perfectly anticipated from the onset (in the absence of SL), to one whose outcome is of a proba-
bilistic nature (i.e., a stochastic game). For the firm, the SL process generates demand uncertainty
because first-period reviews generate an ex ante probabilistic shift in the second-period demand
curve. For the consumers, the SL process generates uncertainty regarding their future assessment
of the product, should they choose to delay their purchasing decision. Crucially, both the firm’s
pricing decisions and the consumers’ adoption decisions are complicated by the fact that the gen-
eration of product information is endogenous to consumers’ adoption decisions; for instance, if no
sales occur in the first period, then no reviews are generated, and therefore nothing is learned by
consumers who delay their purchasing decision.
Under either pricing regime, we show that, conditional on the firm’s first-period announcement,
the equilibrium in the pricing-purchasing game is unique. To distill the effects of SL on the game
2 Pre-announced dynamic pricing is commonly employed in practice indirectly; for instance, firms may set a regularprice and offer introductory price-cuts (e.g., via promotional offers or coupons) or charge introductory price-premiums(e.g., for cinema/theater premieres). Note also that maintaining a constant price is a special case of a pre-announcedprice plan. On the other hand, responsive pricing (sometimes referred to as “contingent pricing”) is commonly observedin online commerce; for example, Amazon.com is known to employ complex dynamic-pricing algorithms (Marketplace2012).
4 Author: Dynamic Pricing with Social Learning
between the firm and consumers, we compare the results of our equilibrium analysis against those
of a benchmark model in which the firm and consumers remain forward-looking, but in which we
“switch off” the SL process (i.e., by cutting off the firm and consumers’ access to product reviews).
This comparison yields three main sets of insights, which we summarize below.
In terms of the interaction of the SL process with consumers’ adoption decisions, we establish
the following results. Under pre-announced pricing, the opportunity to learn from product reviews
in the second period increases consumers’ tendency to delay their purchasing decision. As a result,
fewer consumers adopt the product in the first period (relative to the absence of SL), verifying the
“free-riding” effect of SL: rather than “gamble” on the performance of the new product, consumers
with relatively low ex ante valuations choose to wait until their peers produce reviews. Under
responsive pricing, we observe the following interesting phenomenon which stems from the firm’s
ability to respond to social learning outcomes by adjusting price: consumers who would have
purchased early in the absence of SL now have an even stronger tendency to do so, while at the same
time consumers who would have delayed their purchase now find it even more appealing to wait.
The end result is quite surprising – the adoption strategy employed by consumers in equilibrium
is identical to that employed in the absence of SL.
In terms of the effects of SL on the firm’s pricing decisions, we identify two significant changes
in the optimal implementation of dynamic pricing. When the firm employs pre-announced pricing,
in the absence of SL it is always optimal to announce a decreasing price path – a practice known
as price-skimming. By contrast, in the presence of SL, the firm finds it optimal to announce an
increasing price plan (this is true unless consumers are highly impatient). The intuition underlying
this phenomenon is associated with the firm’s desire to manage strategic purchasing delays more
effectively (by making consumers “pay” for future information), while at the same time extracting
high rents in favorable SL scenarios (through the high second-period price). When the firm employs
responsive pricing, the first-period price is lower in the presence of SL, while the second-period
price is an ex ante random variable. The lower introductory price aims to entice more consumers to
purchase early, thereby increasing the potency of the SL process, and allowing the firm to capitalize
on its second-period pricing flexibility more effectively. In terms of the structure of the equilibrium
price path, we show that the price may either rise or decline in time, with the latter being the ex
ante more likely outcome of the game.
Our analysis also indicates that, despite the presence of strategic purchasing delays and the fact
that strategic behavior may be amplified by the SL process (i.e., under pre-announced pricing),
the overall impact of SL on expected firm profit is always positive. The beneficial effects of SL have
been previously established for cases in which consumers are assumed to be non-strategic (e.g.,
Ifrach et al. 2013); our work generalizes this finding to the case of strategic consumers.
Author: Dynamic Pricing with Social Learning 5
Our third insight pertains to which class of policies is preferred by the firm when facing strategic
consumers. In the existing literature, a general finding is that responsive pricing, despite its inherent
flexibility, can be sub-optimal for the firm owing to the interplay between the product’s price
path and the purchasing decisions of the forward-looking consumers (e.g., Aviv and Pazgal 2008,
Tang 2006). Our benchmark model concurs with the optimality of pre-announced pricing policies.
However, once we introduce SL into the model, our analytical results and numerical experiments
indicate that this observation is reversed: in the presence of SL, the firm will prefer a responsive
price plan (this is true unless consumers are highly patient and product reviews are uninformative).
Furthermore, we find that the presence of SL has the beneficial effect of aligning the firm and
customers’ preferences for the class of policies that is employed in equilibrium by the firm: in the
absence of SL the firm prefers a pre-announced price plan while consumers prefer a responsive
price plan; in the presence of SL, a responsive price scheme is preferred by both. As a result, SL
ensures that the class of policy selected in equilibrium by the firm is that which achieves higher
total welfare.
2. Literature Review
The literature that considers strategic consumer behavior typically assumes that firms employ one
of two classes of dynamic pricing policies, either (i) pre-announced or (ii) responsive pricing. For
early work focusing on the implications of each of the two classes of policies, the reader is referred
to Stokey (1979) and Landsberger and Meilijson (1985) for pre-announced pricing, and to Besanko
and Winston (1990) for responsive pricing. Since then, both classes of policies have been used
extensively to study various operational decisions; recent examples include Yin et al. (2009), who
assume a pre-announced price plan to study the effects of alternative inventory display formats
on firm profit, and Cachon and Swinney (2009), who assume a responsive price plan to study the
firm’s quantity and salvage-pricing decisions. This paper is a first attempt towards understanding
the relative effectiveness of pre-announced and responsive pricing when the firm and consumers
face quality uncertainty and operate in the presence of SL. As such, our model and analysis are
much in the spirit of Landsberger and Meilijson (1985) and Besanko and Winston (1990), in that
our focus is on highlighting the effects of SL within a simple model of the interactions between the
firm and the consumer population. For each class of policies, we find that the implications of SL
are non-obvious, in terms of both consumer behavior as well as firm pricing decisions.
A question of particular interest in our work is which class of policies (i.e., pre-announced or
responsive) is preferred by the firm. Responsive price plans generate value because they allow
the firm to react optimally to updated information (e.g., demand forecasts, leftover inventory; see
Elmaghraby and Keskinocak (2003)). However, when consumers are forward-looking, responsive
6 Author: Dynamic Pricing with Social Learning
pricing may also have adverse effects owing to the interplay between the product’s price path and
consumers’ adoption decisions, as epitomized by the well-known Coase conjecture (Coase 1972). In
fact, the general consensus in the literature is that a firm facing strategic consumers will prefer a
pre-announced policy (see Cachon and Swinney (2009) for a notable exception). In a multi-period
fixed-quantity setting, Dasu and Tong (2010) provide an upper bound for expected revenues under
pre-announced and responsive pricing schemes, and observe that a pre-announced price plan with
a small number of price changes performs nearly optimally. In a newsvendor model with strategic
consumers, Su and Zhang (2008) argue that an endogenous salvage price (i.e., responsive pricing)
amplifies consumers’ incentive to delay their purchase until the salvage period. Aviv and Pazgal
(2008) compare pre-announced and responsive discounts under a more detailed consumer arrival
process and find that the firm will generally prefer a pre-announced strategy. In the aforementioned
papers, it is assumed that consumers are socially isolated, or equivalently, that no reason exists
for consumer interactions to be relevant (e.g., if product quality is known with certainty, then SL
as described in this paper is clearly redundant). The model we develop agrees with the consensus
(i.e., that pre-announced pricing is optimal for the firm) for the benchmark case in which the firm
and consumers operate in the absence of SL. However, interestingly, we find that the equation
changes dramatically when SL is accounted for: in the majority of cases, the firm’s preference will
be reversed from a pre-announced price plan (in the absence of SL) to a responsive price plan (in
the presence of SL).
Apart from its contribution to the strategic consumer literature, this paper also adds to a grow-
ing stream of literature on “social operations management,” which investigates the implications
of social interactions between consumers for profit-maximizing firms. Hu et al. (2013) consider
a firm selling two substitutable products to a stream of consumers who arrive sequentially and
whose purchasing decisions can be influenced by earlier purchases. Candogan et al. (2012) and
Hu and Wang (2013) study optimal pricing strategies in social networks with positive externali-
ties. Tereyagoglu and Veeraraghavan (2012) consider a setting in which consumers may use their
purchases to display their social status. The type of social interaction considered in this paper is
different: here, consumers interact with each other through buyer reviews with the goal of learning
the unobservable quality of a new product; in this respect, our work connects to the SL literature,
which we discuss next.
In the SL literature, customers are generally assumed to arrive at the firm sequentially and
make once-and-for-all purchasing decisions; in other words, this work typically does not account for
strategic consumer behavior. The seminal papers by Banerjee (1992) and Bikhchandani et al. (1992)
illustrate that when the actions (e.g., adoption decisions) of the first few agents (e.g., consumers)
reveal their private information regarding some unobservable state of the world (e.g., product
Author: Dynamic Pricing with Social Learning 7
quality), subsequent consumers may disregard their own private information and simply mimic
the decision of their predecessor. Bose et al. (2006, 2008) illustrate how a monopolist employing
dynamic pricing can use its pricing decision to control the amount of information that can be
inferred by future consumers from the purchasing decision of the current consumer. Perhaps more
relevant to the post-Internet era are models in which SL occurs based on reviews which reveal ex
post consumer experiences (as is the case in our model), rather than actions which reveal ex ante
private information. Ifrach et al. (2013) study monopoly pricing when consumers report whether
their ex post derived utility was positive or negative. Papanastasiou et al. (2013) focus on the
implications of SL from buyer reviews on the quantity released by a monopolist during a new
product’s launch phase. Bergemann and Valimaki (1997) analyze the diffusion of a new product
in a duopolistic market when consumers and the firm learn the product’s uncertain value from the
experiences of previous product adopters. Importantly, the above work does not account for the
fact that consumers may initially decide not to purchase the product for strategic reasons (i.e., in
order to gain information from product reviews), knowing that they can revisit their decision at a
later point in time. By contrast, YU et al. (2013b) allow for such consumer behavior and investigate
the firm’s optimal responsive pricing strategy. Their equilibrium analysis suggests that the firm
may in fact be worse off in the presence of SL, due to the effects of strategic consumer behavior.
The focus of our paper is more on highlighting the relative differences between pre-announced and
responsive pricing. Furthermore, our approach to modeling the SL process differs substantially and,
as a result, so do our insights, even for the particular case of responsive pricing (see §3 and §5.3
for discussions); for instance, we find that despite strategic consumer behavior, the firm is always
(ex ante) better off in the presence of SL.
Finally, we note that consumers in our model face uncertainty regarding the intrinsic quality of a
new and innovative product but are fully informed about their idiosyncratic preferences. However,
consumers in other settings may initially be uninformed about their idiosyncratic preferences for a
specific product, and learn these preferences over time (e.g., a traveller may initially be uncertain
about his preferences for a ticket on a specific date of travel, but become informed as the date
approaches). Since each consumer’s preferences may depend on various exogenous factors, extant
work has often assumed that this type of uncertainty is resolved exogenously in time. DeGraba
(1995) demonstrates that a monopolist may use supply shortages to induce a buying frenzy among
uninformed consumers (see also Courty and Nasiry (2013) for a dynamic model of frenzies). Swinney
(2011) finds that when consumers learn their preferences over time, the value of quick-response
production practices is generally diminished as a result of forward-looking consumer behavior.
YU et al. (2013a) and Prasad et al. (2011) investigate whether and how retailers should employ
advance selling to uninformed consumers. An exception to the exogenous revelation of preferences
8 Author: Dynamic Pricing with Social Learning
assumed in the above papers is Jing (2011), who considers a responsive pricing problem in which
consumers are more likely to become informed about their preferences as the number of early
adopters increases. The latter paper shares a common feature with our work, in that the generation
of information is endogenous to the adoption decisions of the consumer population; however, the
setting we consider differs significantly (quality uncertainty faced by both consumers and firm, as
opposed to preference uncertainty faced only by consumers), and therefore so do our model insights
(details in §5.3).
3. Model Description
We consider a single firm selling a new product of ex ante uncertain quality, over two periods.
The market consists of a continuum of consumers with total mass M normalized to one, and
each customer demands at most one unit of the product during the course of the selling season.
Customer i’s gross utility from purchasing the product comprises two components: a preference
component, xi, and a quality component, qi (e.g., Villas-Boas 2004, Li and Hitt 2008). The value
of the preference component xi reflects the customer’s idiosyncratic preferences over the product’s
ex ante observable characteristics (e.g., brand, color). We assume that preference components,
xi, are distributed in the population according to the uniform distribution U [0,1]. (The uniform
assumption has no significant bearing on our results, but simplifies analysis and exposition.) The
quality component qi represents the product’s quality for customer i, which is ex ante uncertain;
customers learn the value of qi only after they purchase and experience the product. We assume
that the distribution of ex post quality perceptions in the population is normal, qi ∼N(q, σ2q), where
q is the product’s unobservable mean quality (henceforth referred to simply as product quality)
and σq captures the degree of heterogeneity in post-purchase quality perceptions (relatively more
“niche” products are typically associated with larger σq; see Sun (2012)).3 The wealth-equivalent
net benefit of purchasing the product for customer i in period t, t ∈ {1,2}, is defined by uit =
δt−1c (xi + qi− pt), where pt is the price of the product in period t and δc (0≤ δc ≤ 1) is a discount
factor that applies to second-period purchases and represents the opportunity cost of first-period
product usage. Parameter δc may also be interpreted as a measure of customers’ patience and
therefore as a measure of how “strategic” consumers are (Cachon and Swinney 2009). Throughout
our analysis, we say that customers are “myopic” when δc = 0, and “infinitely patient” when δc = 1.
The product’s unobservable quality, q, is the object of social learning (SL). We assume a sym-
metric informational structure between the firm and consumers: both parties share a common
3 We assume that an individual customer’s xi and qi components are conditionally independent for simplicity inexposition. Such dependance can be incorporated in our model without changing our model insights: ex post qualityperceptions (and therefore product reviews) will be biased by the idiosyncratic preferences of the reviewers, however,rational Bayesian social learners can readily account for this bias provided knowledge of the preference distribution(see Papanastasiou et al. 2013).
Author: Dynamic Pricing with Social Learning 9
and public prior belief over q.4 This belief is expressed in our model through the Normal random
variable qp, qp ∼N(qp, σ2p), where we fix qp = 0 without loss of generality. All customers who pur-
chase the product in the first period report their ex post derived product quality, qi, to the rest
of the market through product reviews (e.g., via an online reviewing platform).5 In the beginning
of the second period, both the firm and consumers observe the reviews of first-period buyers and
update their common belief over the product’s mean quality from qp to qu according to Bayes’
rule.6 Specifically, if a mass of n1 customers purchase and review the product in the first period,
and the average rating of these reviews is R, then the updated belief, qu, is normally distributed,
qu ∼N(qu, σ2u), with mean
qu =n1γ
n1γ+ 1R, where γ =
σ2p
σ2q
. (1)
(The variance of the posterior belief is given by σ2u =
σ2p
n1γ+1.) The posterior mean qu is a weighted
average between the prior mean qp = 0 and the average rating from first-period reviews R. The
weight placed by consumers on R increases with the mass of reviews n1 (henceforth referred to
more naturally as the “number of reviews”) and with the ratio γ.7 Intuitively, a larger number of
reviews renders the average rating more credible. The ratio γ is a measure of the degree of ex ante
quality uncertainty relative to the uncertainty (noise) in individual product reviews. Notice that
when γ = 0, the SL process is essentially inactive: the updated belief, qu, is identical to the prior
belief, qp. This case reflects situations in which SL is either (i) irrelevant, because there is no ex
ante quality uncertainty (i.e., σp→ 0) and therefore nothing to be learned from product reviews,
4 Since the firm and consumers hold the same prior belief, firm actions in our model cannot convey any additionalinformation on product quality to the consumers (i.e., there is no scope for signalling); this informational structureis commonly assumed in the SL literature to focus attention on the peer-to-peer learning process (e.g., Bergemannand Valimaki 1997, Bose et al. 2006, Bose et al. 2008, YU et al. 2013b). Furthermore, although we do not modelexpert/critic reviews explicitly, these may take part in forming the public prior belief; Dellarocas et al. (2007) findthat there is generally little overlap between the informational content of critic reviews and that of consumer reviews.
5 It makes no difference in our model whether consumers report directly on quality, net or gross utility. To see why,note that the product’s price history and the distribution of preferences in the population are common knowledge.Therefore, rational consumers can still employ (1) to learn product quality (as explained subsequently in the maintext), albeit with a simple adjustment performed on the observed average rating R. Furthermore, we may also assumethat only a fraction of first-period buyers produce reviews; this has no qualitative bearing on our model insights.
6 An alternative approach to modeling the SL process has been proposed by Bergemann and Valimaki (1997) (andmore recently adopted in YU et al. (2013b)). In that approach, the authors “assume that a statistic on the aggregateperformance of the new product” is made available after early purchases, which takes the form of an aggregate signalwhose density function is specified by the modeler. Consumers use the realization of the aggregate signal to updatetheir prior, via Bayes’ rule. While that approach has qualitatively similar properties to the one used in our analysis,the processes by which reviews are generated and then aggregated into a signal is unclear. By contrast, in our modelconsumers simply report their own derived quality, and consumers remaining in the market learn directly from thesereports, much as SL occurs in real-world settings.
7 Note that the normalization of the total mass M of consumers to one is inconsequential: for the subsequent analysis
to hold for a general mass M , simply redefine γ as γ = Mσ2p
σ2q
and consider n1 to represent the proportion of the
market that purchases in the first period.
10 Author: Dynamic Pricing with Social Learning
or (ii) useless, because buyer reviews carry no useful information on product quality for future
consumers (i.e., σq→+∞). At the other extreme, when γ→+∞, the SL process dominates the
updated belief: any positive number of buyer reviews causes the firm and consumers to completely
abandon their prior. Throughout our analysis, we refer to γ as the SL influence parameter, since
larger γ effectively means that the SL process is more influential in shaping the quality perceptions
of future consumers.
All of the aforementioned are common knowledge. In addition, each customer has private knowl-
edge of her idiosyncratic preference component, xi. In the beginning of the selling season, the firm
announces either (a) both the first- and second-period prices p1 and p2 (pre-announced pricing),
or (b) only the first-period price p1, with the second-period price p2 to be set in the beginning of
the second period (responsive pricing).8 Consumers exhibit forward-looking behavior: they observe
the firm’s announcement and purchase the product in the first period only if the following two
conditions hold simultaneously: (i) their expected utility from purchase in the first period is non-
negative, and (ii) their expected utility from purchase in the first period is greater than the expected
utility of delaying their purchasing decision. Any customers remaining in the market in the second
period purchase a unit provided their expected utility from doing so is positive. The firm seeks
to maximize its overall expected profit. For simplicity, in our analysis we assume a firm discount
factor of δf = 1; our model insights hold qualitatively for any δf ≥ δc. Furthermore, we normalize
production costs to zero and assume that the firm operates in the absence of any binding capacity
constraints.9
4. Benchmark Case: Forward-Looking Consumers without SocialLearning
We begin our analysis by considering a benchmark model in which the firm and consumers operate
in the absence of SL. Given our model setup, the absence of SL is operationally equivalent to the
case γ = 0, since in this case consumers’ belief over quality remains unaltered throughout the selling
season (see (1)). Results presented in this section can be found in existing literature and are stated
here in our model’s notation for completeness and ease of reference in the subsequent analysis.
8 It is beyond the scope of our analysis to model how the firm credibly commits to prices; rather, our goal is toinvestigate the relative merits of committing to a price path, assuming that such a commitment is feasible (e.g.,through repeated interactions with consumers; see Gilbert and Klemperer (2000), Liu and van Ryzin (2008)).
9 While we make this assumption for simplicity, we also note that it is especially relevant for the case of digitalproducts (e.g., digital music, movies and books, computer software, smart-phone apps); such products have virtuallyzero marginal production costs and are unlimited in availability by their very nature.
Author: Dynamic Pricing with Social Learning 11
4.1. Benchmark: Pre-Announced Pricing
Under pre-announced pricing, the firm announces a price path {p1, p2} from the beginning of the
selling season. Consumers take this announcement as given and respond by deciding whether and
when to buy the product.
Consider first an individual’s best response to an arbitrary pre-announced price plan {p1, p2}.
Customer i’s expected utility from purchase in the first (second) period is E[ui1] = xi + qp − p1
(E[ui2] = xi + qu − p2). Moreover, in the absence of SL, the belief over product quality remains
unchanged throughout the selling season so that qp = qu = 0. Therefore, customer i chooses to
purchase the product in the first period under the two conditions (i) E[ui1] = xi− p1 ≥ 0 and (ii)
E[ui1] ≥ δc(xi − p2) = E[ui2] (see §3). Otherwise, the customer chooses to delay her purchasing
decision until the second period, at which time she purchases a unit if E[ui2]≥ 0.
It is straightforward to show that for any given pre-announced price plan {p1, p2}, there exists
a unique optimal purchasing policy for strategic consumers, such that customer i purchases the
product in the first period only if xi ≥ τ(p1, p2), where
τ(p1, p2) =
p1 if p1 ≤ p2,p1−δcp2
1−δc if p1 > p2 and p1− δcp2 ≤ 1− δc,1 if p1 > p2 and p1− δcp2 > 1− δc.
(2)
When the product has an increasing or constant price plan (p1 ≤ p2), any customer with non-
negative first-period expected utility purchases the product, since a delayed purchase incurs a
higher (or equal) cost and discounted utility. When the price of the product is decreasing (p1 > p2),
two possible cases arise. If the first-period price is sufficiently close to the second-period price
(p1 − δcp2 ≤ 1 − δc), customers with relatively higher expected utility will purchase in the first
period in order to avoid discounted second-period utility, while customers with relatively lower
expected utility will delay their purchase until the second period. Otherwise, if the second-period
price is significantly lower than the first-period price (p1−δcp2 > 1−δc), no customer will purchase
the product in the first period. Furthermore, if customer i does not purchase in the first period,
then she purchases in the second period provided xi ≥ p2.
Given knowledge of consumers’ response to any arbitrary price plan, the firm chooses {p∗1, p∗2}
to maximize its overall profit, given by
πbp(p1, p2) = p1[1− τ(p1, p2)] + p2[τ(p1, p2)− p2]+,
where we have used the notation [r]+ = max[r,0]. The firm’s optimal pricing policy, which (along
with the consumers’ adoption policy, as described above) specifies the unique equilibrium of the
pricing-adoption game, is described in Proposition 1 (see also Landsberger and Meilijson (1985)).
12 Author: Dynamic Pricing with Social Learning
Proposition 1. Under pre-announced pricing, the firm’s unique optimal pricing policy is a
decreasing price plan with p∗1 = 2(1−δc)4−(1+δc)2
and p∗2 = 1+δc2p∗1. Moreover, p∗1 (p∗2) is strictly decreasing
(increasing) in δc, while firm profit πbp(p∗1, p∗2) is strictly decreasing in δc.
All proofs are relegated to the Appedix. The firm chooses a decreasing price plan and exercises price
skimming, but only to the extent permitted by forward-looking consumer behavior. Furthermore,
as customers become “more strategic” (i.e., more patient), prices approach each other and firm
profit decreases.
4.2. Benchmark: Responsive Pricing
Next, consider the case of responsive pricing. In the first period, the firm announces the first-period
price and strategic consumers respond by making first-period purchasing decisions. In the second
period, the firm observes the valuations of consumers remaining in the market and announces
the second-period price; consumers remaining in the market respond by making second-period
purchasing decisions.
Consider first the equilibrium of the second-period subgame. As shown in Besanko and Winston
(1990), for any first-period price p1, consumers in the first period adopt a threshold purchasing
policy; therefore, consumers remaining in the market in the second period have total mass x and
idiosyncratic preference components xi distributed uniformly U [0, x], for some x ∈ [0,1]. Given
a second-period price p2, consumers remaining in the market purchase the product if E[ui2] =
δc(xi − p2) ≥ 0. The firm chooses the second-period price p∗2 to maximize πbr2(p2) = p2(x − p2).
Therefore, given any x, there exists a unique equilibrium in the second-period subgame played
between the firm and consumers. Specifically, the firm sets the second-period price at p∗2(x) = x2,
and customers remaining in the market purchase the product provided xi > p∗2.
In the first period, the firm and consumers anticipate the effects of their actions on the equilibrium
of the second-period subgame. Given a first-period price p1, consumer i forms beliefs (which are
correct in equilibrium) over x and p∗2(x) and purchases only if (i) E[ui1] = xi − p1 ≥ 0 and (ii)
E[ui1] ≥ δc(xi − p∗2(x)) = E[ui2]. More specifically, given any first-period price p1, there exists a
unique optimal purchasing strategy for strategic consumers, such that customer i purchases the
product in the first period only if xi ≥ χ(p1), where
χ(p1) =
{1 if p1 > 1− δc
2,
p1
1− δc2if p1 ≤ 1− δc
2.
(3)
When the product’s introductory price is sufficiently high (p1 > 1 − δc2
), all customers prefer
to delay their purchase until the second period, knowing that the firm will lower the price of
the product significantly in order to maximize its second-period profit. When this is not the case
Author: Dynamic Pricing with Social Learning 13
(p1 ≤ 1− δc2
), high-expected-utility customers purchase in the first period, while relatively lower-
expected-utility customers prefer to purchase at the lower second-period price.
The firm anticipates customers’ first-period response to any arbitrary price p1, as well as the
outcome of the second-period subgame, and chooses the introductory price p∗1 to maximize its
overall profit, given by
πbr(p1) = p1[1−χ(p1)] +χ(p1)2
4.
The firm’s optimal first-period pricing policy, which specifies the unique subgame-perfect equilib-
rium of the pricing-purchasing game, is described in Proposition 2 (see also Besanko and Winston
(1990), but note that in their analysis the firm is assumed to discount at the same rate as the
consumers).
Proposition 2. Under contingent pricing, the firm’s unique optimal first-period pricing policy
is p∗1 = 24a−a2 , where a= 1
1− δc2. Moreover, the equilibrium price path satisfies p∗1 > p
∗2 and πbr(p
∗1) is
strictly decreasing in δc.
In equilibrium, the product’s price always declines over time, and as customers become more
strategic the firm’s profit decreases.
5. Forward-Looking Consumers with Social Learning
How does the introduction of SL in the above benchmark model change the consumers’ adoption
strategy and the firm’s pricing decisions? In this section, we consider the game between the firm and
consumers when consumers are forward-looking and interact socially through product reviews. We
analyze the game under pre-announced and responsive pricing policies in turn, before performing
a comparison between the two classes of policies. To facilitate exposition of our results, we begin
by illustrating how the SL process manifests in firm and consumer decision-making.
5.1. A Rational Belief over Social Learning
The firm and consumers observe the reviews of first-period buyers and use them to refine their
belief over the product’s quality, q, from qp to qu. In the second period of our model, customers
base their purchasing decisions on qu. For instance, if customer i remains in the market for the
second period and the product’s price is p2, then she will purchase the product if and only if
E[ui2] = xi+ qu−p2 ≥ 0 (recall that qu is the mean of the updated belief qu; see (1)). Now consider
customer i’s first-period decision: in order for her to make a decision on whether to purchase the
product or delay her purchasing decision, she must form a rational belief over her second-period
expected utility. In turn, to achieve the latter it is necessary for her to form a rational belief over
the posterior parameter qu; that is, the posterior mean qu is viewed in the first period as a random
variable, which is realized after the reviews of first-period buyers have been observed by customer
i. This rational belief, termed the “pre-posterior” distribution of qu, is described in Lemma 1.
14 Author: Dynamic Pricing with Social Learning
Lemma 1. Suppose that n1 product reviews are available to customers remaining in the market
in the second period. Then the pre-posterior distribution of qu has a Normal density function with
mean zero (i.e., equal to qp) and standard deviation σp√
n1γn1γ+1
.
Reassuringly, ex ante, product reviews have no effect, on average, on the mean of customers’
belief – had this not been the case, then the prior belief held by consumers would be inconsistent.
The standard deviation of the pre-posterior distribution (which measures the extent to which the
posterior mean is likely to depart from the prior mean) depends on the amount of information
made available to the customer through product reviews, and includes uncertainty regarding both
the product’s quality q, as well as the noise in individual buyers’ product reviews. Perhaps counter-
intuitively, as the number of reviews increases and the information conveyed through these reviews
becomes more precise, the variance of the pre-posterior distribution increases. To see why this is
the case, note that if n1 = 0, then no additional information is available in the second period, and
customers’ posterior mean, qu, is exactly equal to the prior mean, i.e., qu = qp = 0. On the other
hand, as the number of reviews increases, the posterior mean is more likely to depart further from
the prior mean, consistent with the pre-posterior distribution having greater variability.
Importantly, in the analysis that follows, the number of reviews generated in the first period
(n1) will be an equilibrium outcome, because it depends directly on customers’ first-period adop-
tion decisions. To conclude this section, we introduce the following notation, which will facilitate
exposition of our results.
Definition 1. The probability density function f(·;z) corresponds to a zero-mean Normal ran-
dom variable of standard deviation σ(z), where σ(z) := σp
√(1−z)γ
(1−z)γ+1. Define also F (·;z) as the
corresponding cumulative distribution function, and let F (·;z) := 1−F (·;z).
Note that f(·;z) as described above is not well-defined for the case of γ = 0. Therefore, in the
subsequent analysis, the absence of SL is represented by the limiting case of γ→ 0.
5.2. Pre-Announced Pricing
In this section, we discuss the equilibrium of the game between the firm and consumers when
the firm adopts a pre-announced pricing policy. The sequence of events is similar to that in §4.1,
augmented to account for the SL process. In the first period, the firm announces the full price path
{p1, p2}. Customers take this announcement as given, and make first-period purchasing decisions.
In the second period, customers observe the reviews of first-period buyers, update their beliefs
over product quality, and then make second-period purchasing decisions. We first characterize
consumers’ first- and second-period decisions under any arbitrary price path; we then analyze the
firm’s pricing problem.
Author: Dynamic Pricing with Social Learning 15
5.2.1. Consumers’ Purchasing Strategy To gain an understanding of the process by which
consumers make purchasing decisions in the presence of SL, it is instructive to first consider how
the actions of individual consumers affect the utility of their peers. In settings characterized by SL,
information on product quality is both generated and consumed by the customer population: each
additional early purchase generates an additional product review, which in turn enables customers
remaining in the market to make an incrementally better-informed purchasing decision. In our
model, any individual consumer’s expected utility from delaying her purchasing decision (i.e., until
the second period) increases with the number of customers who choose to purchase the product
early (i.e., in the first period).
Lemma 2. For any given pre-announced price plan {p1, p2}, each customer’s second-period
expected utility is strictly increasing in the number of first-period buyers.
That is, given a price path {p1, p2}, a better-informed future decision is associated in our model
with a higher future expected utility. The equilibrium purchasing strategy adopted by consumers
is one characterized by a form of free-riding, since customers are enticed to wait for the informa-
tion generated by others rather than experiment with the new product themselves. However, this
tendency to delay is mediated by the endogenous generation of information: the larger the number
of customers who strategically delay their purchase, the less well-informed future decisions will be.
Theorem 1. For any given pre-announced price plan {p1, p2}, there exists a unique equilibrium
in the purchasing game played between consumers. Specifically:
- In the first period, customer i purchases the product if and only if xi ≥ θ(p1, p2), where
θ(p1, p2) =
{1 if p1− δcp2 > 1− δc,y if p1− δcp2 ≤ 1− δc,
and y ∈ (p1,1) is the unique solution to the implicit equation
y− p1 = δc
∫ +∞
p2−y(y+ qu− p2)f(qu;y)dqu. (4)
- In the second period, customer i purchases the product if and only if p2 − qu ≤ xi < θ(p1, p2),
where qu is the realized posterior mean belief over quality.
When the first-period price is significantly higher than the second-period price (p1−δcp2 > 1−δc),
we observe what is referred to as “adoption inertia”: the significant cost benefit associated with
second-period purchases makes all customers choose to defer their purchasing decision, even though
second-period decisions will be made without any additional information from product reviews
(since no sales occur in the first period). On the contrary, when p1 is not much higher than p2,
a positive number of customers purchase the product in the first period. The left-hand side of
16 Author: Dynamic Pricing with Social Learning
(4) represents the marginal customer’s first-period expected utility from purchase, while the right-
hand side represents her expected utility from delaying the purchasing decision. Notice that the
lower limit of the integral accounts for the fact that, after observing the reviews of her peers, the
customer will only purchase the product if her updated expected utility is positive – delaying the
purchasing decision grants customers the right, but not the obligation, to purchase in the second
period.
By examining the properties of the solution to (4), we may deduce the following properties of
the purchasing equilibrium.
Corollary 1. The number of first-period buyers, 1− θ, is
- decreasing (increasing) in the first-period price p1 (second-period price p2),
- decreasing in customers’ discount factor, δc,
- decreasing in the degree of SL influence, γ.
Since in the presence of SL customers have an added informational incentive to delay their purchas-
ing decisions, we observe a larger number of strategic delays than in the absence of SL (i.e., γ→ 0).
An increase in the strength of the SL influence (γ) results in an increase in strategic purchasing
delays – as SL becomes more influential, consumers become “more strategic.”
5.2.2. Firm’s Pricing Policy For the firm, optimizing the pre-announced price plan is a con-
voluted task owing to the intricate interaction between its pricing decisions, the adoption decisions
of strategic consumers, and the ex ante uncertain effects of the SL process on the valuations of
second-period consumers. The results of §5.2.1 generate several intriguing questions. How should
the firm adjust its pricing decisions to accommodate the SL process in the presence of strategic
consumers? Should it even allow for SL to occur, or should it induce adoption inertia (Theorem
1)? Moreover, given that SL amplifies strategic consumer behavior (Corollary 1), is the firm better
or worse off in its presence?
Given knowledge of customers’ response to any arbitrary pre-announced price plan, the firm
chooses {p∗1, p∗2} to maximize its expected profit, defined by
πp(p1, p2) = p1[1− θ] + p2
(∫ p2
p2−θ[θ+ qu− p2]f(qu;θ)dqu +
∫ +∞
p2
θf(qu;θ)dqu
), (5)
where the dependence of the threshold θ on p1 and p2 has been suppressed to simplify notation. The
first and second terms correspond to first- and second-period profit, respectively. While first-period
profit is deterministic, the firm’s second-period profit is ex ante uncertain owing to the demand
uncertainty generated by the SL process – depending on the realization of the posterior parameter
qu, either none (low qu) or a fraction (moderate qu) or all (high qu) of the remaining customers
purchase the product in the second period.
Author: Dynamic Pricing with Social Learning 17
To gain insight into the firm’s problem, we begin with the following result, which allows us to
subsequently restrict our attention to policies that generate a strictly positive number of first-period
sales and a strictly positive number of expected second-period sales.
Proposition 3. Under pre-announced pricing, it can never be optimal for the firm to induce
adoption inertia in the first period, or to pre-announce a second-period market exit.
Note that both adoption inertia (i.e., announcing {p1, p2} such that p1− δcp2 > 1− δc) as well as a
pre-announced second-period exit from the market (i.e., announcing {p1, p2} with p2→+∞) confine
all sales to take place in one of the two selling periods; such outcomes are strictly sub-optimal for
the firm. Therefore, the significance of Proposition 3 is to establish that the SL process will always
be “active” in equilibrium; comparing with Theorem 1, the result implies that a solution to the
firm’s problem exists and satisfies p∗1− δcp∗2 ≤ 1− δc.
Now let us consider the firm’s optimal pricing policy in more detail. To this end, the optimality
conditions of problem (5) are, unfortunately, not very informative. To illustrate the effects of SL
on the firm’s pre-announced price plan, we decompose the overall impact of SL into its two main
effects, and analyze the implications of each effect in turn.
1. The behavioral effect changes consumers’ purchasing behavior in the first period: as γ increases,
consumers’ informational incentive to delay their purchase increases, resulting in a larger
number of strategic purchasing delays.
2. The valuation effect shifts the demand curve faced by the firm in the second period: depending
on whether (and the extent to which) reviews are favorable or not, the firm faces a population
of relatively higher or lower valuations for the product in the second period.
Consider first the implications of the behavioral effect, ignoring for now the valuation effect (that
is, suppose for now that the valuations of consumers remaining in the market do not change in
the second period). The behavioral effect of SL simply makes consumers more patient; therefore,
we may leverage the result of our benchmark model in Proposition 1 and consider what happens
when the firm faces consumers who are relatively more patient. This allows us to assert that the
behavioral effect of SL causes a decrease in p∗1, an increase in p∗2, and a decrease in expected firm
profit.
Next, we turn to the implications of the valuation effect, which are less straightforward. In
particular, note that at the time of the firm’s announcement (i.e., in the first period), the impact
of the valuation effect is uncertain because it depends on the realization of the posterior parameter
qu. Since this effect operates on the valuations of second-period consumers, we are predominantly
interested in how it affects the firm’s pre-announced second-period price. To illustrate, we construct
a paradigm in which we “switch off” the behavioral effect of SL by making customers myopic, but
18 Author: Dynamic Pricing with Social Learning
maintain the valuation effect by retaining the influence of first-period reviews on second-period
consumer valuations.
Lemma 3. Suppose that δc = 0, and fix the first-period price at some arbitrary p1. Then for any
k > 0, p∗2|γ=k > p∗2|γ→0. Furthermore, π∗p|γ=k >π
∗p|γ→0, where π∗p = πp(p1, p
∗2).
All else being equal, the firm will choose a higher second-period price in the presence of SL, and
extract higher overall expected profit. The rationale behind Lemma 3 is based on the symmetric
nature of the uncertainty faced by the firm (i.e., qu is viewed ex ante as a Normal random variable;
see Lemma 1). For every favorable SL scenario (there exists a continuum of these), there exists a
corresponding unfavorable “mirror” scenario that is equally probable. By increasing the second-
period price, the firm is able to capitalize on highly favorable scenarios more effectively, while at
the same time its profit in the corresponding highly unfavorable scenarios is at worst zero.
How do the behavioral and valuation effects combine? In terms of the optimal price plan, we have
seen that the behavioral effect causes a decrease in p∗1 and an increase in p∗2, while the valuation
effect causes a further increase in p∗2. The following special case of our model suggests that the
overall impact of SL on the optimal price plan can be significant.
Proposition 4. Suppose δc = 1. Then for any γ > 0, the optimal price plan satisfies p∗1 < p∗2.
Recall from Proposition 1 that in the absence of SL, an increasing price plan can never be opti-
mal. Surprisingly, although Proposition 4 applies only for the extreme case of infinitely patient
consumers, the optimality of an increasing price plan in the presence of SL is the rule rather than
the exception. To illustrate the generality of this result, we present the numerical experiments of
Figure 1, in which we vary the two main parameters in our model, namely, customers’ degree of
patience (δc) and the SL influence parameter (γ). Observe that the firm maintains a decreasing
price path (which is optimal in the absence of SL) only when δc is very low or γ is extremely low
(i.e., SL is absent or uninformative).
Recall that δc can be viewed as a measure of the opportunity-cost perceived by consumers of
postponing their purchase to the second period. In practical settings, δc can be low (i.e., high
perceived opportunity-cost), for example, when customers attach value to the “newness” of the
product (e.g., being the first to watch a new Broadway show). In such cases, our analysis indicates
that the firm finds it most profitable to charge a price-premium in the first period aimed towards
extracting rents from high-valuation consumers, followed by a lower second-period price.
By contrast, when consumers are not highly impatient, the firm announces a low “introductory”
price followed by a higher “regular” price. Here, we find that it is optimal for the firm to pre-
announce a second-period information-premium; that is, to charge consumers for the privilege of
Author: Dynamic Pricing with Social Learning 19
0.30
0.40
0.50
0.60
0.70
0.80
0.90
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
pri
ce
δc
first-period price
second-period price
first-period price
second-period price
0.45
0.55
0.65
0.75
0.85
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
pri
ce
γ
first period
second period
𝑝1∗, 𝛾 = 1
𝑝2∗ , 𝛾 = 1
𝑝1∗, 𝛾 = 0
𝑝2∗ , 𝛾 = 0
𝑝1∗
𝑝2∗
Figure 1 Optimal first- and second- period prices for different combinations of parameter values. Default
parameter values: δc = 0.85, σp = 1.
making a better-informed purchasing decision. This premium has two effects. First, it counter-
balances consumers’ increased willingness-to-wait in anticipation of product reviews, and shifts
demand back to the first period. Second, from those consumers who choose to wait despite the
high second-period price, the firm extracts high profit in cases of highly favorable SL scenarios.
Crucially, notice that the first effect feeds forward and reinforces the second, in the sense that
a larger number of first-period reviews (generated by shifting demand back to the first period)
renders highly favorable SL scenarios ex ante more probable (by increasing the ex ante variability
of qu; see Lemma 1).10
Even though the firm can do its best to mitigate the negative effects of SL on strategic consumer
behavior through its pricing policy, it remains unclear whether the overall impact of SL on expected
firm profit is positive or negative. When consumers are myopic (δc = 0), Lemma 4 is sufficient
to establish that the SL process is ex ante beneficial for the firm. However, the picture is not as
clear when consumers are strategic (δc > 0). In particular, SL exacerbates the amount of strategic
purchasing delays, which causes a relative decrease in firm profit. On the contrary, the valuation
effect generates a relative increase in firm profit. Thus, the overall impact of SL depends on the
relative magnitude of these two opposing effects. Proposition 5 takes consumer strategicness to the
extreme.
Proposition 5. Suppose that δc = 1. Then the firm achieves greater expected profit in the pres-
ence of SL (γ > 0) than it achieves in its absence (γ→ 0).
Interestingly, even for the case in which consumers are infinitely patient, the presence of SL is
ex ante beneficial for the firm. While the corresponding result is difficult to obtain analytically
10 Swinney (2011) illustrates than when consumers’ preferences are revealed exogenously over time (i.e., as opposed toproduct quality being learned endogenously through SL), it is optimal for the firm to employ an increasing price planso as to decrease strategic purchasing delays among uninformed consumers. The increasing price plan in our modelalso reduces strategic delays but, more importantly, serves the purpose of reinforcing the (endogenous) second-periodvaluation effect.
20 Author: Dynamic Pricing with Social Learning
for more general cases of our model, since the positive valuation effect dominates the negative
behavioral effect when δc = 1, it is perhaps not surprising that the result generalizes across our
numerical experiments (see Figure 2).
0.25
0.27
0.29
0.31
0.33
0.35
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
πp
δc
gamma=1
gamma=0
0.25
0.26
0.27
0.28
0.29
0.3
0.31
0.32
0.33
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
πp
γ
delta_c=0.4
delta_c=0.85
𝛾 = 1 𝛾 = 0
𝛿𝑐 = 0.4 𝛿𝑐 = 0.85
Figure 2 Optimal pre-announced profit at different combinations of γ and δc. Parameter values: σp = 1.
To conclude this section, it is instructive to consider how the result of Proposition 5 is affected
if we restrict the firm to charge a constant price (i.e., by adding the constraint p1 = p2 to problem
(5)). This issue is of particular relevance in settings in which fairness considerations are important
for long-term firm-customer relationships (e.g., The New York Times 2007), or when implementing
price changes is costly or impractical (see also Aviv and Pazgal (2008)) – in such settings, the firm
may be reluctant to price intertemporally. We find that the result (i.e., that SL is beneficial for
the firm in the presence of strategic consumer behavior) continues to hold even when the firm is
restricted to maintaining a constant price (see Appendix, proof of Proposition 5).
5.3. Responsive Pricing
Now suppose that the firm does not commit to a second-period price from the beginning of the
selling season. Under a responsive pricing strategy, the game between the firm and consumers is
modified as follows: In the beginning of the first period, the firm sets the first-period price, p1, and
consumers make first-period purchasing decisions. In the beginning of the second-period, the firm
and consumers observe the product reviews generated by first-period buyers and update their belief
over product quality. The firm then sets the second-period price, p2, and consumers remaining in the
market make second-period purchasing decisions. The two-period stochastic game between the firm
and consumers is analyzed in reverse chronological order; we seek a subgame-perfect equilibrium.
5.3.1. Second-Period Pricing and Purchasing Strategies In order to analyze the second-
period subgame, we temporarily assume that in the first period, for any first-period price p1 chosen
by the firm, customers adopt a threshold purchasing policy; the validity of this assumption is
proven in the next section.
Author: Dynamic Pricing with Social Learning 21
At the beginning of the second period and as a result of customers’ first-period purchasing
decisions, the firm faces a population of consumers of total mass x with idiosyncratic preference
components xi distributed uniformly U [0, x], for some x satisfying 0≤ x≤ 1. Moreover, both the
firm and consumers have access to the product reviews of first-period buyers, and arrive at a
posterior mean qu, according to Bayes’ rule in (1). At an arbitrary second-period price p2, customer
i’s expected utility from purchase in the second period is E[ui2] = δc(xi+ qu−p2). The firm’s profit
in the second period is defined by
π2(p2) = p2 [min (x, x+ qu− p2)]+,
and the unique equilibrium of the second-period subgame is characterized in Proposition 6.
Proposition 6. Under responsive pricing, given any qu and x, there exists a unique equilibrium
in the second-period subgame played between the firm and consumers. Specifically:
- The firm’s optimal second-period pricing policy is defined by
p∗2(qu, x) =
qu if qu > x,qu+x
2if − x < qu ≤ x,
0 if qu ≤−x.
- Customer i purchases the product in the second period if and only if p∗2(qu, x)− qu ≤ xi < x.
In the second period, customers simply choose to purchase the product if their expected utility from
purchase (given what they have learned from product reviews and the firm’s decision p∗2) is non-
negative. The firm’s profit-maximizing p2 depends on the outcome of SL, qu, as well as customers’
first-period purchasing decisions which specify x (that is, qu and x jointly define the demand curve
faced by the firm in the second period). If qu is very high (a sign of high quality for firm and
consumers), the firm finds it most profitable to choose the market-clearing price p∗2 = qu. When qu
is at intermediate levels, the firm chooses a price at which only a fraction of consumers remaining
in the market choose to adopt the product. Finally, when qu is very low, the firm cannot extract
profit at any positive price p2, and therefore exits the market; this is signified by a second-period
price of p∗2 = 0.
5.3.2. First-Period Pricing and Purchasing Strategies In the first period, the firm and
consumers anticipate the effects of their actions on the equilibrium of the second-period subgame.
However, note that since the realization of the posterior mean qu is ex ante uncertain, the equilib-
rium in the second-period subgame is itself uncertain; unlike the benchmark case in §4.2, the firm
and consumers in this case form rational probabilistic beliefs over the second-period equilibrium.
We first consider the consumer’s first-period decision for any arbitrary price p1 and subsequently
address the firm’s first-period pricing problem. Recall that customer i purchases the product in
22 Author: Dynamic Pricing with Social Learning
the first period only if her expected utility from a first-period purchase is higher than that of a
delayed purchasing decision.
Theorem 2. Under responsive pricing and any given first-period price p1, there exists a unique
optimal first-period purchasing strategy for consumers. Specifically, customer i purchases the prod-
uct in the first period if and only if xi ≥ ζ(p1), where
ζ(p1) =
{1 if p1 > 1− δc
2,
p1
1− δc2if p1 ≤ 1− δc
2.
Customers’ first period adoption strategy under responsive pricing is reminiscent of that under pre-
announced pricing. If the first-period price is relatively high (p1 > 1− δc2
), all consumers delay their
purchasing decision (adoption inertia) in anticipation of a significantly lower second-period price.
If the first-period price is not too high (p1 ≤ 1− δc2
), customers with relatively higher valuations
purchase in the first period, while the rest of the population delays the purchasing decision (case
ζ = p1
1− δc2).
Theorem 2 also suggests the following, quite remarkable phenomenon.
Corollary 2. Under responsive pricing and for any given first-period price p1, the first-period
purchasing strategy adopted by forward-looking consumers is independent of the SL influence γ.
Surprisingly, consumers make identical first-period purchasing decisions in the absence (γ→ 0) and
in the presence of SL (γ > 0). When the firm maintains the flexibility to respond to the reviews
of first-period buyers by adjusting the product’s price, it is no longer the case that SL generates
a relative increase in strategic purchasing delays in the first period; that is, SL has no behavioral
effect, in contrast with the case of pre-announced pricing.11
More specifically, what occurs is that the presence of SL does not change the decisions of the
forward-looking consumers, instead, it strengthens the incentives that drive these decisions. To
illustrate, define the “value of SL” for customer i, vsl(xi), as the difference between the customer’s
equilibrium expected utility from delaying her decision in the presence (i.e., for some γ > 0) and
absence of SL (γ → 0); that is, vsl(xi) := E[u∗i2]|γ>0 − E[u∗i2]|γ→0. It is straightforward to show
that, at any p1, we have vsl(xi)> 0 for xi ≤ ζ(p1) (i.e., for customers who delay) and vsl(xi)< 0
for xi > ζ(p1) (i.e., for customers who buy early); Figure 3 provides a numerical illustration of
this “rotational” effect. Interestingly, this implies that high-xi customers have an even stronger
preference for purchasing early in the presence of SL, for if they do not, the flexible second-period
firm will leave them with relatively less expected surplus in the second period (as compared to that
in the absence of SL).
11 Although YU et al. (2013b) study a setting similar to that considered in this section, the sharp characterizationof consumers’ first-period strategy in Theorem 2, and in particular how this strategy compares to the case of no SL(Proposition 2), appears to be novel; these characterizations are possible in our model owing to our different approachto modelling the review-generation and SL processes (see §3 and footnote 6).
Author: Dynamic Pricing with Social Learning 23
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
xi
E[u_i1]
E[u_i2] no SL
E[u_i1] with SL
𝐸 𝑢𝑖1∗
𝐸 𝑢𝑖2∗ , 𝛾 = 0
𝐸 𝑢𝑖2∗ , 𝛾 = 1
𝑥𝑖
Figure 3 Equilibrium expected utility from purchasing early, E[u∗i1], and delaying (with and without SL), E[u∗i2],
for customer type xi. Both in the presence of SL (γ = 1) and in its absence (γ→ 0), the indifferent consumer has
xi = 0.39. Consumers with xi > 0.39 (xi < 0.39) have relatively lower (higher) expected utility from delaying their
decision in the presence of SL. Parameter values: p1 = 0.25, δc = 0.7, σp = 1.
Consider next the firm’s first-period pricing decision. While SL has no effect on the first-period
purchasing decisions of strategic consumers, the same is not true for the firm’s first-period pricing
strategy. In the presence of SL and strategic consumers, the firm chooses p∗1 to maximize its overall
expected profit, which is defined by
πr(p1) = p1[1− ζ] +
∫ ζ
−ζ
(qu + ζ
2
)2
f(qu; ζ)dqu +
∫ +∞
ζ
ζquf(qu; ζ)dqu, (6)
where the dependence of ζ on p1 has been suppressed to simplify notation. As opposed to the
case of pre-announced pricing, problem (6) accounts for the fact that in the second period, the
firm will adjust the product’s price in response to the posterior realization of qu. According to
the results of Theorem 2 and Proposition 6, any first-period price chosen by the firm generates
a unique subgame-perfect equilibrium in the pricing-purchasing game. Although it is difficult to
characterize the firm’s optimal first-period price analytically, the following proposition reduces the
problem to a simple one-dimensional optimization problem over a finite interval.
Proposition 7. Under responsive pricing, it can never be optimal for the firm to induce adop-
tion inertia in the first period.
Similarly as in the case of pre-announced pricing, Proposition 7 suggests that the firm will never
induce adoption inertia under responsive pricing, and that the SL process will always be “active”
in equilibrium. Furthermore, combined with Theorem 2, the result of Proposition 7 implies that
an optimal first-period price p∗1 exists and satisfies 0≤ p∗1 ≤ 1− δc2
.12
12 This result contrasts with Jing (2011), who shows that when consumers learn their preferences through SL (andthe firm faces no ex ante uncertainty), adoption inertia may arise in equilibrium since the firm may find it preferableto sell to an uninformed and therefore more homogeneous population in the second period. Since consumers in ourmodel know their preferences ex ante (and therefore consumer heterogeneity is not affected by the SL process), suchincentives do not arise.
24 Author: Dynamic Pricing with Social Learning
Given that the firm will always avoid adoption inertia, how does the first-period price in the
presence of SL compare to that in its absence? Observe in Figure 4 that the firm will introduce the
product at a relatively lower price in the presence of SL. (This occurs in general in our experiments,
unless γ is extremely high.) The rationale underlying this strategy is associated with the firm’s
preference for high ex ante variability in qu (i.e., high variance in the pre-posterior distribution of
qu). To understand this preference, note that the firm’s second-period profit is convex increasing in
qu (see (6)); this implies that, ceteris paribus, the firm’s expected profit is increasing in the ex ante
variability of qu. Next, recall from Corollary 2 that the presence of SL does not change consumers’
first-period adoption strategy. Therefore, the relatively lower introductory price is an attempt to
reinforce the SL process (i.e., to increase the variance of the pre-posterior distribution of qu: see
Lemma 1) by increasing the number of product reviews generated in the first period. Of course,
this tendency to lower p1 is tempered by the firm’s desire to (a) extract substantial profit in the
first period and (b) keep a substantial portion of the consumer population in the market so that
it may capitalize on its second-period pricing flexibility.
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
pri
ce
δc
Series1
Series3
0.38
0.40
0.42
0.44
0.46
0.48
0.50
0.52
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
pri
ce
γ
Series1𝑝1∗, 𝛾 = 1
𝑝1∗, 𝛾 = 0
𝑝1∗
Figure 4 Optimal first-period price in the presence and absence of social learning, for different combinations of
parameter values. Default parameter values: δc = 0.85, σp = γ = 1.
Next, we consider the structure of the equilibrium price path. Recall that in the absence of SL
(γ→ 0), the equilibrium price path is always decreasing (see Proposition 2). In the presence of SL,
given p∗1, the optimal second-period price p∗2 depends on the firm and consumers’ posterior mean
belief qu (see Proposition 6 and replace x with ζ(p∗1)). Although customers remaining in the market
in the second period have relatively lower preference components xi than those who purchase in
the first period, it is no longer the case that the price of the product will always be lower in the
second period. In particular,
Corollary 3. The ex ante probability of an increasing price path satisfies 0<P (p∗1 < p∗2)< 1
2.
Author: Dynamic Pricing with Social Learning 25
An increasing price path may indeed arise in equilibrium; however, a decreasing price path is ex
ante more likely. To see why, note that if consumers’ valuations did not change at all (i.e., as would
be the case in the absence of SL), then the firm would charge a price lower than p∗1 in the second
period (see Proposition 2). It follows that for p∗2 to be higher than p∗1 in the presence of SL, it must
be the case that consumers’ valuations have increased considerably in the second-period; given the
ex ante symmetric nature of SL, the ex ante probability of this occurring cannot be greater than
one half. This result suggests a contrast between the structure of the price path under responsive
and pre-announced pricing: when the firm commits to a price path ex ante, this path will generally
be increasing (unless customers are highly impatient); when the firm does not commit to a price
path, it is more likely that prices will decline over time.
To conclude this section, we present the following result, which characterizes how the firm’s
expected profit is affected by changes in the SL influence (γ) as well as the ex ante quality uncer-
tainty surrounding the product (σp, keeping γ fixed; see Lemma 1).
Proposition 8. Under responsive pricing, the firm’s expected profit is strictly increasing in both
the SL influence (γ), and in quality uncertainty (σp).
The intuition underlying Proposition 8 is again based on the firm’s preference for high ex ante
variability in qu, as discussed previously. We note that the beneficial effects of SL on expected
profit when the firm adjusts prices dynamically have been established previously in the literature,
but under the assumption that consumers are non-strategic (e.g., Bose et al. 2006, Ifrach et al.
2013). Proposition 8 generalizes this finding to the case of forward-looking consumers.13
6. Pre-Announced vs. Responsive Pricing
A recurring theme in the recent literature that considers strategic consumer behavior is the value of
price-commitment for the firm. For instance, Aviv and Pazgal (2008) report that when customers
are forward-looking (and in the absence of future rationing risk), pre-announced pricing is a more
effective way of managing strategic consumer behavior than responsive pricing, in the sense that it
allows the firm to extract higher overall profit (see “announced” and “contingent” pricing in their
model). Our benchmark model replicates this prediction.
Lemma 4. In the absence of SL (γ→ 0) and for any δc ∈ (0,1), firm profit under pre-announced
pricing is strictly higher than that under responsive pricing.
13 Note that the firm in our model is by assumption more patient than the consumer population (since it does notdiscount second-period profit). In the analysis of YU et al. (2013b), the authors also allow for the opposite case.Interestingly, they provide a numerical example in which they observe that when the consumer population is morepatient than the firm, reviews that are relatively more informative may generate a relative decrease in expected firmprofit. From the proof of Proposition 8 it becomes evident that such cases do not arise in our model, even if we allowfor an arbitrary combination of firm and consumer patience. In other words, the result of the proposition does notdepend on the fact that the firm discounts at a lower rate than consumers.
26 Author: Dynamic Pricing with Social Learning
The question of interest in this section is whether price-commitment is preferred by the firm when
SL is accounted for. Surprisingly, we observe that, in general, the opposite is true. Let us begin by
first proving the sub-optimality (in terms of expected firm profit) of a pre-announced price plan in
the presence of SL for the special case of myopic consumers.
Lemma 5. Suppose that δc = 0. Then responsive pricing generates higher firm profit than pre-
announced pricing for any γ > 0.
When customers are myopic but second-period customers’ valuations are affected by the reviews of
their peers, it is always beneficial for the firm to maintain price-flexibility. To see why, notice that
any arbitrary first-period price generates the same amount of first-period sales (and profit) under
either pricing policy; however, under responsive pricing the firm can maximize its second-period
profit by choosing the best price given the content of product reviews.
The sub-optimality of price-commitment in the presence of SL generalizes far beyond the special
case of myopic consumers considered in Lemma 5. Although a comparison between the two policies
is difficult to perform analytically for general cases of our model, our numerical study points to
three main observations (see Figure 5): first, a responsive pricing policy is optimal in the vast
majority of cases; second, the cases in which a pre-announced price plan is preferred are those that
combine patient customers with weak SL influence (i.e., high δc and low γ); third, in those cases
in which a pre-announced price plan is optimal, the increase in profit with respect to the optimal
responsive price plan is much smaller (1.5% on average) than the corresponding increase in profit
when a responsive price plan is optimal (8.7% on average).
0.25
0.27
0.29
0.31
0.33
0 0.5 1 1.5 2
exp
ecte
d p
rofi
t
γ
pre-announced
responsive
𝛿𝑐 = 0.8
γ γ
0.5 1 1.5 2
0.2
0.4
0.6
0.8
1
Figure 5 Left: Region plot for the optimal class of dynamic pricing policy as a function of consumer patience
(δc) and SL influence (γ); shaded regions mark the optimality of a responsive pricing policy; white regions mark
the optimality of a pre-announced policy. Parameter values: σp = 1.
Right: Expected firm profit under the optimal pre-announced and responsive pricing policies as a function of SL
precision (γ), for the case of δc = 0.8. Parameter values: σp = 1.
Author: Dynamic Pricing with Social Learning 27
The rationale underlying these observations is as follows. The flexibility offered by a responsive
pricing policy is significantly advantageous for the firm when the valuations of second-period con-
sumers are likely to change significantly as a result of SL. Ceteris paribus, a significant change
in consumer valuations occurs when the number of first-period reviews is large (high n1) and/or
when the influence of SL is strong (high γ). When the value of γ is low, the only way in which the
firm can profit substantially from its second-period flexibility is if it generates a very large number
of first-period sales/reviews. As customers become progressively more patient (i.e., δc increases)
the firm is required to introduce the product at a progressively lower price in order to achieve the
volume of reviews necessary to capitalize on its flexibility. It is for this reason that, when δc is high
and γ is low, the firm prefers to employ the optimal pre-announced price path, rather than the
optimal responsive price plan (which would entail either a large amount of sales at a very low price,
or reduced effectiveness of pricing flexibility). Moreover, we note that even when the firm does
prefer a pre-announced price plan, the advantage with respect to a responsive policy is generally
very small and quickly disappears as γ increases from zero (see, e.g., the right-hand side plot of
Figure 5).
Next, consider the customers’ perspective. In the absence of SL, the firm’s profit is maximized
under a pre-announced price plan (Lemma 4), but it is straightforward to show that the opposite
is true for consumer surplus.
Lemma 6. In the absence of SL (γ → 0) and for any δc ∈ (0,1), consumer surplus is greater
under responsive pricing that it is under pre-announced pricing.
In the absence of SL, a planner seeking to maximize consumer surplus by simply choosing the class
of policy to be employed by the firm would opt for a responsive price plan. In our numerical study,
the result of Lemma 6 continues to hold when SL is accounted for (i.e., for all γ > 0). Combining this
observation with Figure 5, we conclude that SL has the positive effect of aligning the preferences of
the firm and consumers regarding which class of policy arises in equilibrium (this is true whenever
responsive pricing is preferred by the firm). Clearly, this has the further implication that, among
the two classes of policies available to the firm, the class chosen in equilibrium is generally the one
that maximizes expected total welfare (i.e., the sum of expected firm profit and expected consumer
surplus).
7. Conclusion
We have presented a stylized analysis of the effects of quality uncertainty and social learning on
the purchasing behavior of strategic consumers, and the pricing decisions of a monopolist firm.
Recent research has highlighted that firms may neglect the ever-increasing sophistication of the
modern-day consumer at their peril. Yet this research has itself neglected perhaps one of the most
28 Author: Dynamic Pricing with Social Learning
important aspects of this sophistication: the ability of consumers to exchange experiences and learn
from their peers. This paper demonstrates that pricing techniques which have come to constitute
conventional wisdom may in fact be overturned by the increasing influence of SL on consumer
decision-making. For instance, we have shown that price-commitment, which has been advocated
as an effective way of managing strategic consumer behavior, is in fact suboptimal for the firm
once SL is accounted for. This suggests that managers pricing products for which SL (i.e., product
reviews) is known to be a significant driver of demand should be encouraged to price products
dynamically in response to buyers’ sentiments (e.g., media items on Amazon.com). Furthermore,
this paradigm shift from pre-announced to responsive pricing constitutes a “win-win” situation –
both the firm and the consumer population benefit from responsive prices.
Another main result of our analysis is that, even taking into account the negative effects of strate-
gic consumer behavior, the SL process is one that should be endorsed and promoted by modern-day
firms (in our model, the SL influence parameter γ and expected firm profit are positively related
under either class of pricing policy). There are at least three dimensions along which firms can act
to enhance the influence of SL on consumers’ product evaluations and purchasing decisions. The
first is to simply encourage the consumer population to pay more attention to buyer opinions; for
instance, KIA and Ford motors have recently invested heavily in advertising campaigns aimed at
promoting consumer attention to buyer reviews (see their “Reviews and Recommendations” and
“Good Reviews” campaigns, respectively). The second is to provide the platform upon which con-
sumers can communicate and exchange their product experiences; examples of high-profile firms
that have actively facilitated the exchange of consumer experiences on their online spaces include
Amazon.com, Dell Computers, and Apple. A third dimension is to increase the precision of buyer-
generated reviews by asking consumers to rate products on multiple dimensions (e.g., see reviews
on Hotels.com) rather than simply providing a uni-dimensional star rating.
A more subtle implication of our work is associated with firms’ new product development (NPD)
strategy. Specifically, we find that in the presence of SL, greater ex ante quality uncertainty is
associated with higher expected firm profit. This result suggests that firms developing products for
which SL is known to be significantly influential (e.g., high-tech electronics) should be encouraged
to take risks and innovate with their new products (as opposed to incrementally improving previous
product versions). When the resulting product proves to be a success, our results suggest that the
SL process will be handsomely rewarding, while in cases of product failures, the corresponding
downside is relatively less severe.
Finally, we have seen that when consumers can anticipate future prices perfectly (e.g., in cases of
price-commitment), the SL process generates a relative increase in strategic purchasing delays; by
contrast, this relative increase is not present when future prices are unpredictable. This may have
Author: Dynamic Pricing with Social Learning 29
implications for researchers and firms attempting to estimate the effects of SL on firm profits (e.g.,
as in Moretti (2011)). Specifically, neglecting to account for consumers’ increased strategicness may
lead to over-estimates of the impact of SL since intertemporal demand shifts (i.e., from earlier to
later in the selling season) may be misdiagnosed as demand generated through the SL process.
Appendix
Proof of Proposition 1 (Outline)
Using the expression for τ(p1, p2) in (2), it is straightforward to verify that for price plans satisfying p1 ≤ p2
profit is bounded by π≤ 12, and that the same is true for price plans satisfying both p1 > p2 and p1− δcp2 >
1 − δc simultaneously. Next, for price plans satisfying p1 > p2 and p1 − δcp2 ≤ 1 − δc, the firm’s problem
is concave and the optimal price plan {p∗1, p∗2} as quoted in the proposition can be derived via first order
conditions, noting also that π(p∗1, p∗2)> 1
2for all δc ∈ (0,1). The properties stated follow readily.
Proof of Proposition 2 (Outline)
Using the expression for χ(p1) in (3) it follows that π(p1) = 14
for p1 > 1− δc2
, because no sales occur in the first
period and the optimal second-period price is 12. Next, for p1 ≤ 1− δc
2we have π(p1) = p1(1− ap1) +
a2p214
=
p1−ap21[1− a
4] where a= 1
1− δc2
and 1<a< 2. The profit function is thus strictly concave in p1 (in this region
of prices) with a unique maximizer at p∗1 = 24a−a2 . Finally note that p∗1 = 2
4a−a2 <1a
= 1− δc2
for all a∈ (1,2)
and the profit function is continuous; this implies that the global maximizer of the profit function is p∗1 and
that at least some sales occur in the first period. The properties of the price path and profit follow readily.
Proof of Lemma 1
In the second period, if the product reviews of n1 first-period buyers have a mean of R, then by Bayes’ rule,
the posterior belief over q, denoted by qu, is Normal with a mean of qu, where
qu =σ2q
n1σ2p +σ2
q
qp +n1σ
2p
n1σ2p +σ2
q
R
=1
n1γ+ 1qp +
n1γ
n1γ+ 1R,
where qp is the mean of the prior belief and γ =σ2p
σ2q.
In the first period, the posterior mean belief qu is viewed as a random variable (r.v.), since it depends
on the unobservable realization of product quality q, as well as the noise in first period reviews (i.e., it is
subject to sampling error). Specifically, if the product’s quality realization is q, the sample mean of n1 (i.i.d.
Normal) reviews, R, follows R∼N(q,σ2q
n1) so that qu | q ∼N
(σ2q
n1σ2p+σ2
qqp + n1γ
n1γ+1q,(
n1γ
n1γ+1
)2 σ2q
n1
). Therefore,
since q is an ex ante Normal r.v., q∼N(qp, σ2p),
E[qu] =E(E[qu | q]) =E
(σ2q
n1σ2p +σ2
q
qp +n1γ
n1γ+ 1q
)= qp, and
Var[qu] =E(Var[qu | q]) + Var(E[qu | q])
=
(n1γ
n1γ+ 1
)2(σ2q
n1
+σ2p
)=
(n1γ
n1γ+ 1
)2(σ2q (n1γ+ 1)
n1
)=
n1γ
n1γ+ 1σ2p .
To complete the proof, recall that we have normalized qp to zero (see §3).
30 Author: Dynamic Pricing with Social Learning
Proof of Lemma 2
Let v2(xi, y) := δc∫ +∞p2−xi
(xi + qu − p2)f(qu;y)dqu denote the second-period expected utility of customer xi,
conditional on n1 = 1 − y customers purchasing in the first period. We will show that ∂v2∂y
< 0. Rewrite
v2(xi, y) = δc∫ +∞p2−xi
(xi + qu− p2)f(qu;y)dqu = δc∫ +∞−∞ mi(qu)f(qu;y)dqu, where mi(qu) = 0 if qu < p2−xi and
mi(qu) = xi + qu − p2 otherwise. Note that mi(qu) is non-negative, convex and increasing in qu. It follows
that the integral∫ +∞−∞ mi(qu)f(qu;y)dqu is strictly increasing in the variance of the pre-posterior distribution
of qu. In turn this variance is decreasing in y. (Proceeding in a similar manner, it is straightforward to show
also that ∂v2∂γ
> 0.)
Proof of Theorem 1
The proof proceeds in three steps. First, we show that any purchasing equilibrium must be characterized by
a first-period threshold policy. Second, we establish the condition under which this threshold is such that no
customers purchase in the first period (adoption inertia). Third, we show that when adoption inertia does
not occur, the implicit equation which characterizes the equilibrium threshold has exactly one solution in
the interval [0,1], and that this solution lies in the open interval (p1,1).
Step 1. Consider any arbitrary price plan {p1, p2}. Define customer i’s best response function given that
(any) ψ customers choose to purchase the product in the first period by bi(ψ) and note that
bi(ψ) =
{1 (buy now) if ∆i(ψ) =E[ui1]−E[ui2]≥ 0
0 (defer decision) if ∆i(ψ) =E[ui1]−E[ui2]< 0
We establish strict monotonicity of ∆i(ψ) in xi, thereby proving that the equilibrium must admit a first-
period threshold structure. We have
∆i(ψ) = xi− p1− δcE[(xi + qu− p2)+
]= xi− p1− δc
∫ +∞
p2−xi
(xi + qu− p2)f(qu; 1−ψ)dqu,
such that qu ∼N(
0,ψσ4
p
ψσ2p+σ2
q
). The derivative with respect to xi is given by
∂∆i
∂xi= 1− δc
∫ +∞
p2−xi
f(qu; 1−ψ)dqu
= 1− δcF (p2−xi; 1−ψ)
= 1− δcΦ(p2−xiσ(1−ψ)
)> 0.
Since the above monotonicity holds for any arbitrary ψ customers, it follows that any pure-strategy equilib-
rium must admit a threshold structure.
Step 2. Given that any equilibrium must follow a threshold structure, we now identify the condition under
which the first-period purchasing threshold is such that no customer purchases in the first period (i.e.,
θ(p1, p2) = 1). To do so, we consider the highest xi customer (i.e., xi = 1). Her first-period expected utility
from purchase is 1− p1. If she delays, the threshold structure of the equilibrium implies that no customer
buys in the first period, no reviews are generated, and her expected utility from purchasing in the second
period is δc(1−p2). Thus, the condition for adoption inertia is simply 1−p1 < δc(1−p2), or p1−δcp2 > 1−δc.
Author: Dynamic Pricing with Social Learning 31
Step 3. If the above condition for adoption inertia does not hold, this implies that at least some customers
will purchase the product in the first period. Denote customer i’s expected utility from purchase in the first
period by v1(xi), v1(xi) = xi − p1, and customer i’s expected utility from purchase in the second period,
conditional on customers with xi ≥ y choosing to purchase the product in the first period, by v2(xi, y),
v2(xi, y) = δc∫ +∞p2−xi
(xi + qu− p2)f(qu;y)dqu.
We will show that the indifference equation v1(y) = v2(y, y) has exactly one solution, y∗, satisfying
y∗ ∈ (p1,1). Notice first that ∂v1∂y
= 1. Next, rewrite v2(y, y) = δc∫ +∞p2−y
(y + qu − p2)f(qu;y)dqu = δc(y −
p2)Φ(y−p2σ(y)
)+ δcσ(y)φ
(y−p2σ(y)
)=: u(y). The derivative of u(y) yields
∂u
∂y=δc(y− p2)φ
(y− p2
σ(y)
)∂
∂y
(y− p2
σ(y)
)+ δcΦ
(y− p2
σ(y)
)+ δcσ
′(y)φ
(y− p2
σ(y)
)+ δcσ(y)φ′
(y− p2
σ(y)
)∂
∂y
(y− p2
σ(y)
)=δc
(Φ
(y− p2
σ(y)
)+σ′(y)φ
(y− p2
σ(y)
)),
and since σ′(y) < 0, we have ∂u∂y< 1. By comparing the derivatives of the two functions, it follows that
v1(y) = v2(y, y) can have at most one solution in the interval y ∈ [0,1]. From Step 2, we know that v1(1)>
v2(1,1). Moreover, note that v1(p1) = 0 and that v2(y, y) > 0 for all y ∈ [0,1]; therefore, we know that
v1(p1)< v2(p1, p1). Combining the above, we deduce that a unique solution y∗ exists and satisfies y∗ ∈ (p1,1).
Proof of Corollary 1
Note first that if p1 − δcp2 > 1 − δc then no customer purchases in the first period, for any parameter
values. If p1 − δcp2 ≤ 1 − δc, then from the proof of Theorem 1, we have θ ∈ (p1,1) and the number of
first-period buyers is 1 − θ. From Theorem, 1, this θ satisfies the equation θ − p1 − u(θ) = 0 (†), where
u(θ) = δc∫ +∞p2−θ
(θ + qu − p2)f(qu;θ)dqu. The properties stated in the proposition can be obtained via the
implicit function theorem.
Taking the total derivative of (†) with respect to p1, we have
∂θ
∂p1
− 1− ∂u
∂θ
∂θ
∂p1
=∂θ
∂p1
(1− ∂u
∂θ
)− 1 = 0.
From the proof of Theorem 1 we know that ∂u∂θ< 1 and therefore the above equation implies ∂θ
∂p1> 0, which in
turn means that the number of first-period buyers is decreasing in p1. Next the total derivative with respect
to p2 yields
∂θ
∂p2
−(∂u
∂p2
+∂u
∂θ
∂θ
∂p2
)=
∂θ
∂p2
−(−δcF (p2− θ;θ) +
∂u
∂θ
∂θ
∂p2
)=
∂θ
∂p2
(1− ∂u
∂θ
)+ δcF (p2− θ;θ) = 0.
Therefore ∂θ∂p2
< 0, which implies that the number of first-period buyers is increasing in p2.
As for the result pertaining to customers’ degree of patience δc, we have
∂θ
∂δc−(∂u
∂δc+∂u
∂θ
∂θ
∂δc
)=∂θ
∂δc
(1− ∂u
∂θ
)− ∂u
∂δc= 0.
Since ∂u∂δc
> 0, it follows that ∂θ∂δc
> 0.
32 Author: Dynamic Pricing with Social Learning
Finally, note that we may rewrite u(θ) = δc(θ− p2)Φ(θ−p2σ(θ)
)+ δcσ(θ)φ
(θ−p2σ(θ)
), from which it follows that
∂u∂σp
= δc∂σ(θ)
∂σpφ(θ−p2σ(θ)
)> 0, since σ(θ) = σp
√(1−θ)γ
(1−θ)γ+1. Taking the total derivative of (†) with respect to σp,
we have
∂θ
∂σp−(∂u
∂σp+∂u
∂θ
∂θ
∂σp
)=
∂θ
∂σp
(1− ∂u
∂θ
)− δc
∂σ(θ)
∂σpφ
(θ− p2
σ(θ)
)= 0.
Therefore ∂θ∂σp
> 0, which implies that the number of first-period buyers is decreasing in σp. In a similar
manner, it can also be verified that ∂θ∂σq
< 0, and since γ =σ2p
σ2q
the two latter results imply that ∂θ∂γ> 0.
Proof of Proposition 3
Note that among all price plans which induce adoption inertia, those which achieve the highest profit have
second-period price p2 = 12
and achieve total profit of π∗in = 14. An equivalent way for the firm to achieve
this profit is by announcing a price plan { 12,+∞}, i.e., to announce market exit in the second period. Under
both policies, exactly half of the population purchases at price 12. To prove that none of these policies can
be optimal, we will show that a specific price plan, which does not induce adoption inertia, achieves profit
strictly greater than 14. Since this price plan is not necessarily optimal for the firm, it follows that neither
adoption inertia nor market exit can be optimal policies.
Consider the price plan { 12, 1
2}. In this case, first-period profit is π1 = 1
2(1− θ) for some θ ∈ ( 1
2,1) (see
Theorem 1). Denoting second-period expected profit by π2, we will show that π1 + π2 >14. Equivalently,
we will show π2 = 12E[s2]> 1
2(θ− 1
2), i.e., E[s2]> (θ− 1
2) where s2 denotes second-period sales. Note that
qu ∼N(0, σ(θ)) and that
s2(qu) =
0 if qu ≤ 1
2− θ,
θ+ qu− 12
if 12− θ < qu ≤ 1
2,
θ if qu >12.
Next, decompose s2(qu) into s2(qu) = sa(qu) + sb(qu), where
sa(qu) =
0 if qu ≤ 1
2− θ,
θ+ qu− 12
if 12− θ < qu ≤ θ− 1
2,
2θ− 1 if qu > θ− 12,
and
sb(qu) =
0 if qu ≤ θ− 1
2,
−θ+ qu + 12
if θ− 12< qu ≤ 1
2,
1− θ if qu >12.
We have
E[s2] =
∫ +∞
−∞s2(qu)f(qu;θ)dqu =
∫ +∞
−∞(sa(qu) + sb(qu))f(qu;θ)dqu (7)
=
∫ +∞
−∞sa(qu)f(qu;θ)dqu +
∫ +∞
−∞sb(qu)f(qu;θ)dqu. (8)
The symmetry of the r.v. qu around zero and the function sa(qu) imply∫ +∞−∞ sa(qu)f(qu;θ)dqu = 2θ−1
2= θ− 1
2.
Moreover, since sb(qu) is a non-negative function of qu (and positive for some values of qu), it follows that∫ +∞−∞ sb(qu)f(qu;θ)dqu > 0. Therefore, we have E[s2]> θ− 1
2, and the proof is complete.
Author: Dynamic Pricing with Social Learning 33
Proof of Lemma 3
We show the pricing and profit impact of the valuation effect in turn, starting from the pricing impact. For
δc = 0, we have θ(p1, p2) = p1 for all p1, p2. Consider some arbitrary p1; denoting by p∗2 the optimal second
period price we want to show that for γ > 0 we have p∗2 >p12
(note that for γ → 0 we have p∗2 = p12
from
Proposition 1). Expected profit is given by
π(p2) = p1(1− p1) + p2
∫ p2
p2−p1
(p1 + q− p2)f(q;p1)dq+ p2
∫ +∞
p2
p1f(q;p1)dq.
The first-order necessary condition for p∗2 is
dπ
dp2
=
∫ p2
p2−p1
(p1 + q− p2)f(q;p1)dq+ p2
(−∫ p2
p2−p1
f(q;p1)dq+ p1f(p2;p1)
)+
∫ +∞
p2
p1f(q;p1)dq− p2p1f(p2;p1)
=p1
∫ +∞
p2−p1
f(q;p1)dq− 2p2
∫ p2
p2−p1
f(q;p1)dq+
∫ p2
p2−p1
qf(q;p1)dq= 0.
Notice that∫ +∞p2−p1
f(q;p1)dq >∫ p2p2−p1
f(q;p1)dq > 0 and∫ p2p2−p1
qf(q;p1)dq > 0. Defining a :=∫ +∞p2−p1
f(q;p1)dq,
b :=∫ p2p2−p1
f(q;p1)dq and c :=∫ p2p2−p1
qf(q;p1)dq, we have
ap1− 2bp2 + c= 0,
where a > b > 0. Next notice that for p2 = p12
(i.e., the optimal second-period price at γ = 0) we have c= 0.
It follows that at p2 = p12
, we have dπdp2
= (a− b)p1 > 0, which implies p∗2 >p12
.
Next, consider the impact of social learning on profit. Note that since we have fixed p1 and customers are
myopic, the firm’s first-period profit is fixed and we need only consider differences in expected second-period
profit. Assume that the firm announces p2 = p12
irrespective of γ. Denoting second period profit at second-
period price p2 by π2(p2), we will show that E[π2(p2)]|γ=k =E[π2(p2)]|γ→0 for any k > 0. Since p2 is optimal
for γ→ 0 but suboptimal for γ = k, this implies that the firm achieves higher expected second-period profit,
and therefore higher overall expected profit, in the presence of SL. The firm’s second-period profit at p2 = p12
as a function of qu is
π2(p2) =
0 if qu ≤− p1
2,(
p214
+ qup12
)if − p1
2< qu ≤ p1
2,
p212
if qu >p12.
Now, note that E[π2(p2)]|γ→0 =p214
and consider the difference E[π2(p2)]|γ=k−E[π2(p2)]|γ→0.
E[π2(p2)]|γ=k−E[π2(p2)]|γ→0 =
∫ p12
− p12
(p2
1
4+ qu
p1
2
)f(qu;p1)dqu +
∫ +∞
p12
p21
2f(qu;p1)dqu−
p21
4. (9)
Replacingp214
=∫ +∞−∞
p214f(qu;p1)dqu and breaking up the integral accordingly, it is straightforward to show
E[π2(p2)]|γ=k − E[π2(p2)]|γ→0 = 0. Since the second-period price p2 = p12
is suboptimal for any γ > 0, we
conclude that π∗|γ=k >π∗|γ→0 for any k > 0.
Proof of Proposition 4
Using the result of Proposition 3, we may restrict our attention to policies {p1, p2} which result in θ ∈(p1,1). The indifference equation which connects the price plan to the first-period purchasing threshold,
θ, is θ − p1 = δc∫ +∞p2−θ
(θ + qu − p2)f(qu;θ) (?). From the proof of Lemma 2, we know that for any γ > 0,
δc∫ +∞p2−θ
(θ+qu−p2)f(qu;θ)> δc(θ−p2); that is, the marginal customer’s second-period expected utility in the
presence of SL (γ > 0) is greater than that in its absence (γ→ 0). Therefore, (?) implies that the equilibrium
price plan and θ satisfy θ−p1 > δc(θ−p2). At the extreme case of δc = 1, the last inequality implies p2 > p1.
34 Author: Dynamic Pricing with Social Learning
Proof of Proposition 5
Denote firm profit at prices {p1, p2} in the absence of SL by πn(p1, p2). From Proposition 1 it follows that
when δc = 1, π∗n = maxp1,p2 πn = 14. In the presence of social learning, the firm can achieve profit equal to π∗n
by inducing adoption inertia (see proof of Proposition 3). But from the result of Proposition 3, we know that
adoption inertia is suboptimal for the firm for any γ > 0 and the result stated in the proposition follows.
(Furthermore, from the proof of Proposition 3 we also know that if the firm chooses the price plan p1 = p2 = 12
it achieves expected profit strictly higher than 14. It follows that the result continues to hold even if the firm
is restricted to announcing a constant price.)
Proof of Proposition 6
In the second period, the firm faces a mass of x customers with valuations uniformly distributed U [qu, qu+ x].
If qu ≤−x then no customer will purchase the product in the second period at any non-negative price. In
this case, the firm exits the market; this is denoted by an optimal price p∗2 = 0, and second-period profit
is π2 = 0. If qu > −x then the firm’s profit function is π2(p2) = p2 min{qu + x− p2, x}. Clearly, any price
p2 < qu cannot be optimal; we therefore restrict our attention to the function π2(p2) = p2(qu + x− p2). If
qu > x, the profit function is decreasing for p2 ≥ qu, and we have p∗2 = qu with associated profits π∗2 = qux. If
−x < qu ≤ x, the profit function is increasing at p2 = qu and concave, and we have p∗2 = qu+x2
with associated
profits π∗2 =(qu+x
2
)2. The customer’s second-period purchasing decision is trivial: she purchases provided
xi + qu− p∗2 is non-negative.
Proof of Theorem 2
We first establish that any pure-strategy purchasing equilibrium must follow a threshold structure. Assume
that (any) ψ customers purchase the product in the first period and that the firm’s profit-maximizing price
in the second period is some p2. In the spirit of the proof of Theorem 1, at first-period price p1, we have
∆i(ψ) = xi− p1− δcE[(xi + qu− p2)+
]= xi− p1− δc
∫ +∞
p2−xi
(xi + qu− p2)f(qu; 1−ψ)dqu,
where qu ∼N(
0,ψσ4
p
ψσ2p+σ2
q
). As in Theorem 1, we have ∂∆i
∂xi> 0, which establishes that any equilibrium must
follow a threshold structure.
Using the optimal second-period price charged by the firm given by Proposition 6, at any first-period price
p1 and assuming that customers with xi ≥ x purchase in the first period, we have for customer i
∆i = xi− p1− δc∫ +x
x−2xi
(xi + qu−
qu + x
2
)f(qu; x)dqu− δc
∫ +∞
+x
xif(qu; x)dqu.
To find the equilibrium x, we solve the system of equations ∆i = 0 and xi = x. Inserting the latter into the
former
∆i = x− p1− δc∫ +x
−x
(x+ qu
2
)f(qu; x)dqu− δc
∫ +∞
+x
xf(qu; x)dqu
= x− p1− δcx
2
(F (−x; x)− F (x; x)
)− δcxF (x; x)
= x− p1− δcx
2
(F (−x; x) + F (x; x)
)= x− p1− δc
x
2= 0,
which leads to x= p1
1− δc2
. Finally, note that if p1
1− δc2
> 1 = maxxi all customers defer their purchasing decision.
Author: Dynamic Pricing with Social Learning 35
Proof of Corollary 2
Follows directly by comparing the equilibrium outcome of Theorem 2 with equation (3).
Proof of Proposition 7
When p1 > 1− δc2
(i.e., in the case of adoption inertia), the firm’s profit in the presence of social learning
is identical to that in its absence, and equal to 14. From Proposition 2 we know that in the benchmark
case, the firm’s optimal price, p1b, satisfies p1b < 1− δc2
which implies π(p1b)|γ→0 >14. We will show that
π(p1b)|γ=k >π(p1b)|γ→0 for any k > 0, which implies that adoption inertia can never be optimal for the firm
in the presence of SL. Since ζ(p1) of Theorem 2 is independent of γ, the firm’s first-period profit is fixed,
and we need only consider differences in expected second-period profit. The firm’s second-period profit as a
function of qu is defined by
π2(qu) =
0 if qu ≤−ζ,(qu+ζ
2
)2if − ζ < qu ≤ ζ,
quζ if qu > ζ.
Note that π2(qu) is continuous, non-negative, strictly increasing and convex in qu. The firm’s expected
second-period profit is
E[π2] =
∫ +∞
−∞π2(qu)f(qu; ζ)dqu.
Since the ex ante variance of qu is is strictly increasing in γ, it follows that E[π2]|γ=k >E[π2]|γ→0 for any
k > 0. Therefore, π(p1b)|γ=k > π(p1b)|γ→0. Since, p1b is not necessarily optimal in the presence of SL but
nevertheless achieves profit higher than inducing adoption inertia, it follows that adoption inertia can never
be optimal for the firm.
Proof of Corollary 3
In Proposition 6, we may set x= ζ(p1) as per Theorem 2. Then, it follows that p∗2 is higher than p1 when
qu+ζ(p1)
2> p1. Since ζ(p1) = p1
1− δc2
, this implies qu >1−δc1− δc
2
p1. The probability with which the latter occurs is
F(
1−δc1− δc
2
p1; ζ(p1))
and since qu is a zero-mean Normal r.v. this probability is strictly less than 12
for any
p1 > 0 and strictly higher than zero.
Proof of Proposition 8
Consider any fixed p1, such that adoption inertia does not occur in accordance with Proposition 7. Since
ζ(p1) is independent of both γ and σp, the firm’s first-period profit is independent of both. From the proof
of Proposition 7, we have that the firm’s second-period profit π2(qu) is continuous, non-negative, strictly
increasing and convex in qu, and the firm’s expected second-period profit is
E[π2] =
∫ +∞
−∞π2(qu)f(qu; ζ)dqu.
Recall that f(·; ζ) is the density function of a zero-mean Normal random variable with standard deviation
σ(ζ) = σp
√(1−ζ)γ
(1−ζ)γ+1. It follows that σ(ζ) is strictly increasing in both γ and σp. Therefore, E[π2] is strictly
increasing in σ(ζ). Since the result holds for any arbitrary first-period price p1, it must be the case that the
firm’s optimal expected profit is strictly increasing in γ, σp.
36 Author: Dynamic Pricing with Social Learning
Proof of Lemma 4
Using the optimal prices stated in Propositions 1 and 2, we may calculate the difference in optimal firm
profit under pre-announced and responsive pricing as ∆π∗b = π∗bp−π∗br =δ2c(1−δc)
4(9−2δ2c−3δc). Thus, we have ∆π∗b > 0
for any δc ∈ (0,1).
Proof of Lemma 5
Assume δc = 0. Let {p∗1, p∗2} be the optimal pre-announced price plan. Consider a responsive price plan
{p1, p2} with p1 = p∗1 and p2 = p∗2 for qu ≥ p∗2−p∗1, and p2 = p∗2−ε for qu < p∗2−p∗1, where ε is small and positive
(this responsive price plan is not necessarily optimal, but suffices for the purposes of our argument). Notice
that the two pricing policies achieve identical first-period profit, identical reviews and identical second-period
profit for any realization of qu such that qu ≥ p∗2 − p∗1. However, the responsive price plan achieves higher
second-period profit for at least some qu < p∗2 − p∗1: under all such scenarios the pre-announced price plan
achieves zero second-period profit, while the responsive price plan achieves positive profit under at least
some scenarios. Since scenarios under which the responsive price plan achieves higher second-period profit
occur with a strictly positive probability (qu is a Normal r.v.), it follows that the optimal responsive policy
outperforms the optimal pre-announced policy for any γ > 0.
Proof of Lemma 6
Using the optimal prices charged by the firm under each pricing policy from Propositions 1 and 2, it is
straightforward to write down the difference between consumer surplus under pre-announced pricing and
under responsive pricing, ∆sb = sbp− sbr, at any δc as
∆sb =−δc
2(δc
3 + 14 δc2− 51 δc + 36
)8(2 δc
2 + 3 δc− 9)2 .
The above difference is negative for any δc ∈ (0,1), implying that consumer surplus is always greater under
responsive pricing.
References
Aviv, Y., A. Pazgal. 2008. Optimal pricing of seasonal products in the presence of forward-looking consumers.
Manufacturing & Service Operations Management 10(3) 339–359.
Banerjee, A. 1992. A simple model of herd behavior. The Quarterly Journal of Economics 107(3) 797–817.
Bergemann, D., J. Valimaki. 1997. Market diffusion with two-sided learning. RAND Journal of Economics
28(4) 773–795.
Besanko, D., W.L. Winston. 1990. Optimal price skimming by a monopolist facing rational consumers.
Management Science 36(05) 555–567.
Besbes, O., I. Lobel. 2013. Intertemporal price discrimination: Structure and computation of optimal policies.
Working Paper, Columbia University .
Bikhchandani, S., D. Hirshleifer, I. Welch. 1992. A theory of fads, fashion, custom, and cultural change as
informational cascades. Journal of political Economy 992–1026.
Author: Dynamic Pricing with Social Learning 37
Bose, S., G. Orosel, M. Ottaviani, L. Vesterlund. 2006. Dynamic monopoly pricing and herding. The RAND
Journal of Economics 37(4) 910–928.
Bose, S., G. Orosel, M. Ottaviani, L. Vesterlund. 2008. Monopoly pricing in the binary herding model.
Economic Theory 37(2) 203–241.
Cachon, G., P. Feldman. 2013. Price commitments with strategic consumers: Why it can be optimal to
discount more frequently... than optimal. Working paper, University of Pennsylvania .
Cachon, G., R. Swinney. 2009. Purchasing, pricing, and quick response in the presence of strategic consumers.
Management Science 55(3) 497–511.
Cachon, G., R. Swinney. 2011. The value of fast fashion: Quick response, enhanced design, and strategic
consumer behavior. Management Science 57(4) 778–795.
Candogan, O., K. Bimpikis, A. Ozdaglar. 2012. Optimal pricing in networks with externalities. Operations
Research 60(4) 883–905.
Chevalier, J., D. Mayzlin. 2006. The effect of word of mouth on sales: Online book reviews. Journal of
Marketing Research 43(3) 345–354.
Coase, R. 1972. Durability and monopoly. Journal of Law and Economics 15(1) 143–149.
Courty, P., J. Nasiry. 2013. Product launches and buying frenzies: A dynamic perspective. Working Paper,
Hong Kong University of Science and Technology .
Dasu, S., C. Tong. 2010. Dynamic pricing when consumers are strategic: Analysis of posted and contingent
pricing schemes. European Journal of Operational Research 204(3) 662–671.
DeGraba, P. 1995. Buying frenzies and seller-induced excess demand. The RAND Journal of Economics
26(2) 331–342.
Dellarocas, C., X. Zhang, N. Awad. 2007. Exploring the value of online product reviews in forecasting sales:
The case of motion pictures. Journal of Interactive marketing 21(4) 23–45.
Elmaghraby, W., P. Keskinocak. 2003. Dynamic pricing in the presence of inventory considerations: Research
overview, current practices, and future directions. Management Science 49(10) 1287–1309.
Gilbert, R., P. Klemperer. 2000. An equilibrium theory of rationing. RAND Journal of Economics 31(1)
1–21.
Hu, M., J. Milner, J. Wu. 2013. Liking and following and the newsvendor: Operations and marketing policies
under social influence. Rotman School of Management, Working Paper .
Hu, M., Z. Wang. 2013. The double-edged sword of social interactions: Revenue management of network
goods under demand uncertainty. Rotman School of Management, Working Paper .
Ifrach, B., C. Maglaras, M. Scarsini. 2013. Bayesian social learning from consumer reviews. Working Paper,
Columbia University .
38 Author: Dynamic Pricing with Social Learning
Jing, B. 2011. Social learning and dynamic pricing of durable goods. Marketing Science 30(5) 851–865.
Landsberger, M., I. Meilijson. 1985. Intertemporal price discrimination and sales strategy under incomplete
information. RAND Journal of Economics 16(3) 424–430.
Li, J., N. Granados, S. Netessine. 2013. Are consumers strategic? structural estimation from the air-travel
industry. Management Science, forthcoming .
Li, X., L.M. Hitt. 2008. Self-selection and information role of online product reviews. Information Systems
Research 19(4) 456–474.
Liu, Q., G. van Ryzin. 2008. Strategic capacity rationing to induce early purchases. Management Science
54(6) 1115–1131.
Marketplace. 2012. The brave new world of dynamic pricing. (Dec. 19th).
Mersereau, A., D. Zhang. 2012. Markdown pricing with unknown fraction of strategic customers. Manufac-
turing & Service Operations Management 14(3) 355–370.
Moretti, E. 2011. Social learning and peer effects in consumption: Evidence from movie sales. The Review
of Economic Studies 78(1) 356–393.
Papanastasiou, Y., N. Bakshi, N. Savva. 2013. Social learning from early buyer reviews: Implications for
new product launch. London Business School, Working Paper .
Prasad, A., K. Stecke, X. Zhao. 2011. Advance selling by a newsvendor retailer. Production and Operations
Management 20(1) 129–142.
Schlosser, J. 2004. Markdown lowdown. Fortune 149(1) 40.
Stokey, N. 1979. Intertemporal price discrimination. The Quarterly Journal of Economics 93(3) 355–371.
Su, X. 2007. Intertemporal pricing with strategic customer behavior. Management Science 53(5) 726–741.
Su, X., F. Zhang. 2008. Strategic customer behavior, commitment, and supply chain performance. Manage-
ment Science 54(10) 1759–1773.
Sun, M. 2012. How does the variance of product ratings matter? Management Science 58(4) 696–707.
Swinney, R. 2011. Selling to strategic consumers when product value is uncertain: The value of matching
supply and demand. Management Science 57(10) 1737–1751.
Tang, C.S. 2006. Robust strategies for mitigating supply chain disruptions. International Journal of Logistics:
Research and Applications 9(1) 33–45.
Tereyagoglu, N., S. Veeraraghavan. 2012. Selling to conspicuous consumers: Pricing, production, and sourcing
decisions. Management Science 58(12) 2168–2189.
The New York Times. 2007. iPhone owners crying foul over price cut. (Sept. 7th).
Villas-Boas, J.M. 2004. Consumer learning, brand loyalty, and competition. Marketing Science 23(1) 134–
145.
Author: Dynamic Pricing with Social Learning 39
Yin, R., Y. Aviv, A. Pazgal, C.S. Tang. 2009. Optimal markdown pricing: Implications of inventory display
formats in the presence of strategic customers. Management Science 55(8) 1391–1408.
YU, M., H. Ahn, R. Kapuscinski. 2013a. Ration capacity in advance selling to signal product quality. Working
paper, Hong Kong University of Science and Technology .
YU, M., L. Debo, R. Kapuscinski. 2013b. Strategic waiting for consumer-generated quality information:
Dynamic pricing of new experience goods. Working Paper, University of Chicago .